Business continuation policy for server consolidation environment

Description of equivalent WO 03102772 A2

BUSINESS CONTINUATION POLICY FOR SERVER CONSOLIDATION ENVIRONMENT

Portions of this patent application contain materials that are subject to copyπght piotection The copyright owner has no objection to the facsimile reproduction by anyone of the patent document, or the patent disclosure, as it appears in the Patent and Trademark Office file or records, but otherwise reserves all copyπght rights whatsoever

BACKGROUND OF THE INVENTION

As the use of open systems grows, the complexity of managing hundreds or thousands of servers becomes an increasingly difficult task In addition, a demand for increased availability of the applications running on the servers presents a challenge Many information technology (IT) managers are working to move from large numbers of small open systems, many running well below their capacities, to a much smaller number of large-scale enterprise servers running at or near their capacities This trend in the IT industry is called "server consolidation "

One early answer to the demand for increased application availability was to provide one-to- one backups for each server running a critical apphcahon When the critical application failed at the primary server, the application was "failed over" (restarted) on the backup server However, this solution was very expensive and wasted resources, as the backup servers sat idle Furthermore, the solution could not handle cascading failure of both the primary and backup servers

Another possible solution is "N+l clustering," where one enterprise-class server provides redundancy for multiple active servers N+l clustering reduces the cost of redundancy for a given set of applications and simplifies the choice of a server for failover, as an application running on a failed server is moved to the one backup server

However, N+l clustering is not a complete answer to the need for increased application availability, particularly in a true server consolidation environment Enterprises require the ability to withstand multiple cascading failures, as well as the ability to take some serveis offline for maintenance while maintaining adequate redundancy in the server cluster Typical cluster management applications provide only limited flexibility in choosing the propei hosts for potentially tens or hundreds of application groups Examples of commercially available cluster management applications include VERITAS^® Global Cluster Manager™, VERITAS^® Cluster Server, Hewlett-Packard^® MC / Service Guard, and Microsoft^® Cluster Server (MSCS)

N-to-N clustering refers to multiple application groups running on multiple servers, with each application group being capable of failing over to different servers in the cluster For example, a four- node cluster of servers could support three critical database instances Upon failure of any of the four nodes, each of the three instances can run on a respective server of the thiee remaining servers, without overloading one of the three remaining servers N-to-N clustenng expands the concept of N+l clusteπng from a "backup system" to a requirement for "backup capacity" within the servers forming the clustei

What is needed is a business continuity policy that enables critical enterprise applications to survive multiple failures by determining suitable systems for starting applications initially, redistributing applications when systems reach an overloaded condition, and restarting failed applications

SUMMARY OF THE INVENTION

The present invention relates to a method, system and computer program product that establish and maintain a business continuity policy in a server consolidation environment Business continuity is ensured by enabling high availability of applications When an application is started, restarted upon failure, or moved due to an overload situation, a system is selected best fulfilling the requirements for running the application These requirements can include application requirements, such as an amount of available capacity to handle the load that will be placed on the system by the application These requirements can further include system requirements, such as honoring a system limit of a number of applications that can be run on a particular system Respective priorities of applications can be used to determine whether a lower-priority application can be moved oi stopped to free resources for running a higher-priority application

In one feature, a method includes identifying a set of systems in a cluster, wherein each system in the set of systems meets a requirement for hosting an application Identifying the set of systems can involve including a selected system in the set when the selected system meets a prerequisite for the first application Identifying the set of systems can involve including a selected system in the set when the application does not exceed a limit for the selected system

When the set of systems is empty, the method further includes using a respective priority for each of the applications for identifying a resource to free, wherein the resource is one of a plurality of resources, and each resource is associated with at least one of the systems Identifying the resource to free may further include using a respective capacity for each of the systems

The method can further include freeing the lesource such that one of the systems associated with the resource meets the requirement for hosting the application The method may include starting the application on the associated system Freeing the resource can include stopping another application that is using the resource, wherein the other application has a lower priority than the pnority of the first application Freeing the resource can include moving another application that is using the resource to another system when the other application has a lower respective priority than a respective priority of the first application The method can further include determining that the application is to be started This determination can be made when detecting that the application failed Another way to make this determination is to compare the priority of the application with each priority for applications currently running on the systems The application is to be started when the respective prioπty of the application is higher than one of the priorities for the applications running on the systems

In another feature of the invention, an apparatus includes an identifying module to identify a set of systems m a cluster, wherein each system in the set of systems meets a requirement for hosting an application The apparatus further includes a priority module to use a respective priority for each of the applications for identifying a resource to free when the set of systems is empty Each resource is associated with at least one of the systems The apparatus can further include modules to implement the features of the method described above

The foregoing is a summary and thus contains, by necessity, simplifications, generalizations and omissions of detail, consequently, those skilled in the art will appreciate that the summary is illustrative only and is not intended to be in any way limiting Other aspects, inventive features, and advantages of the present invention, as defined solely by the claims, will become apparent in the non- hmiting detailed description set forth below

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention may be better understood, and its numerous objects, features and advantages made apparent to those skilled in the art by referencing the accompanying drawings

Fig 1 provides an example of an environment in which the management system and framework of the present invention operates

Fig 2 shows an example of a cluster configuration in a high-availabihty storage area network

Fig 3 is a flowchart of a method for implementing a business continuity policy in a server consolidation environment

Fig 4 is a flowchart of the Determine Set of Eligible Systems to Host Application Group X step of the flowchart of Fig 3

Fig 5 is a flowchart of the Can Sufficient Capacity and Resources be Freed to Accommodate Application Group X decision point of the flowchart of Fig 3

Figs 6 through 16 show example configurations and failure scenarios handled by the method and system of the present invention Fig 6 shows the calculation of available capacity for a cluster of servers in a server consolidation environment

Fig 7 shows the movement of an application upon failure of one of the servers of Fig 6 and the resulting available capacity in the cluster

Fig 8 shows the movement of another application in the failure scenario of Fig 7

Fig 9 shows an example configuration of database applications in the cluster of Fig 6

Fig 10 shows movement of database applications in a failure scenario in the configuration of Fig 9

Fig 1 1 shows an example of managing application groups using limits and prerequisites

Fig 12 shows a failure scenario in which an application group cannot be failed over

Fig 13 shows stopping a lower-priority application group to free sufficient lesources to enable a higher-pπoπty application to remain available

Fig 14 shows another failure scenario for the configuration of Figs 12 and 13

Fig 15 shows movement of a lower-priority application group to free sufficient resources to enable a higher-priority application group to remain available

Fig 16 shows movement of the higher-priority application group to use the resources freed as a result of the action shown in Fig 15

Fig 17 is a block diagram illustrating a computer system suitable for implementing embodiments of the present invention

The use of the same reference symbols in different drawings indicates similar or identical items While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the Drawings and are described herein in detail It should be understood, however, that the Drawings and Detailed Description are not intended to limit the invention to the particular form disclosed On the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present invention as defined by the appended Claims DETAILED DESCRIPTION

For a thorough understanding of the subject invention, refer to the following Detailed Description, including the appended Claims, in connection with the above-described Drawings Although the present invention is described in connection with several embodiments, the invention is not intended to be limited to the specific forms set forth herein On the contrary, it is intended to cover such alternatives, modifications, and equivalents as can be reasonably included within the scope of the invention as defined by the appended Claims

In the following descπption, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the invention It will be apparent, however, to one skilled in the art that the invention can be practiced without these specific details

References in the specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments Moreover, vaπous features are descπbed which may be exhibited by some embodiments and not by others Similarly, various requirements are described which may be requirements for some embodiments but not other embodiments

Introduction The present invention provides a business continuity policy that proactively determines the best possible system, typically a server in a cluster of servers, to host an application during startup, upon an overload condition, or following an application or server fault The terms server and system are used interchangeably herein, as one of skill in the art will recognize that the present invention also applies to systems operating outside a client/server environment

Fig 1 provides an example of an environment in which the management system and framework of the present invention operates Nodes 1 10A and 1 1 OB at Mountain View (MV) site 130A and nodes 1 IOC and 1 10D at United Kingdom (UK) site 130B are shown for purposes of illustration The invention is not limited to minimum or maximum numbers of nodes and/or sites While typically the term "site" describes a collection of nodes concentrated at a data center or on a campus such that cables can interconnect the nodes and storage devices, geographic concentration is not a requirement for a site A site can include one or moie clusters of nodes and can be viewed as a virtual collection of one or more clusters

MV site 130A and UK site 130B are shown as connected via network 102, which typically corresponds to a private wide area network or a public disti lbution network such as the Internet Common management console 104 is shown to be used for managing nodes and clusters of nodes, although a common management console is not necessary for operation of the invention

Cluster 120A includes nodes 1 10A and HOB at MV site 130A, which are connected via redundant cluster connections 1 15AB-1 and 1 15AB-2 Although only one cluster is shown at MV site 130A, any number of clusters may be included at a site Node 1 10A shares common storage 140A with node 1 10B Node 1 10A is interconnected with storage 140A via interconnection 112A, and node 1 1 OB is interconnected with storage 140A via interconnection 1 12B

Similarly, cluster 120B includes nodes 1 IOC and 110D at UK site 130B, which are connected via redundant cluster connections 115CD-1 and 1 15CD-2 Node 1 IOC shares common storage 140B with node 1 10D Node 1 IOC is interconnected with storage 140B via interconnection 112C and node 1 10D is interconnected with storage 140B via interconnection 1 12D

Fig 2 shows an example of a cluster configured for high availability in a storage area network Cluster servers 210A and 21 OB are configured as servers for the same application program and serve as fail over targets for each other Redundant interconnections 216A and 216B can be redundant heartbeat private network connections via crossover cables between redundant network interface cards (NICs) when two nodes form the cluster When more than two nodes form the cluster, the private network connection can use a hub The private network enables fail over software to recognize when a system oi process has failed Each of clusters 210A and 21 OB has redundant public network connections, such as public network connections 242 A and 244A for cluster server 21 OA and public network connections 242B and 244B for cluster server 210B, to communicate via a public network 240 such as the Internet

Cluster server 210A has redundant connections to a fibre channel storage area network via fibre channel connection 212A to fibre switch 220A and via fibre channel connection 214A to fibre switch 220B Similarly, cluster server 210B is connected to the fibre channel storage area network via fibre channel connection 212B to fibre switch 220B and via fibre channel connection 214B to fibre switch 220A

The fibre channel storage area network provides access by cluster servers 210A and 210B to each of shared storage arrays 230A and 230B Storage arrays 230A and 230B may correspond, for example, to fibre channel RAID arrays Fibre switch 220A is connected to stoiage array 230A via fibre channel connection 222A and to storage array 230B via fibre channel connection 224A Similarly, fibre switch 220B is connected to storage array 230B via fibre channel connection 222B and to storage array 230A via fibre channel connection 224B Redundant connections from the cluster server to the switch and from the switch to the storage array ensure that each of cluster servers 210A and 210B has a connection to a collection of storage devices on the fibre channel network Redundant power sources (not shown) also can be included to provide a backup power source in the event of a power failure Management

To ensure disaster recovery, data loss must be prevented and consistent data maintained even if hardware or software failures occur Data for a particular application should not be allowed to enter a state in which the failuie of the network or a node would leave that application and corresponding application data in an inconsistent or unusable state

Cluster management applications enable administrators to manage multiple, discrete clusters from a single application By coordinating events and actions across clusters, cluster management applications provide a useful tool for managing disaster recovery For example, a second cluster may take over an application running on a primary cluster when no node within the primary cluster can run the application Examples of commercially available clustei management applications include

VERITAS^® Global Cluster Manager™, Hewlett-Packard^® MC / Service Guard, and Microsoft^® Cluster Server (MSCS)

In some cluster management applications, a process called the site master at each site may connect to one or more site slave processes within the site The site master collects all information about all of the clusters and nodes in that site In addition, each site master may connect to all other site masters in the distributed system to share mformation so all site masters have information about the entire distributed system While it is not a requirement that each site have its own master for operation of the invention, a master must have detailed information, sometimes at the software process level, about the state of hardware and software resources at the site The term master refers to a site master and is also referred to heiein as a master process

Typically, a cluster management application constantly monitors the state of software applications in multiple clusters and can determine if an entire site becomes unavailable, such that no node in the clusters at the site is available to run the software application The cluster management application may start the software application at a secondary site unaffected by the circumstances that made the primary site unavailable A cluster management application may be controlled by a user via a user interface, or the clustei management application may be configured to act automatically

In the event that the primary data center is destroyed, the application data must be immediately available at another site, and the application must be immediately started at the other site This level of availability requires lephcation of the data from the primary site to the other site Various data replication applications are available for replicating data across sites, including VERITAS^® Volume Replicator™ ( VVR), Symmetπx Remote Data Facility (SRDF^®) by EMC^® Corporation, Hitachi^® Asynchronous Remote Copy (HARC), Sybase^® Replication, and Continuous Access by Hewlett-Packard^® Factors included in the determination of the "best" server to initially start or to re-start an application include server capacity and finite resource availability In one embodiment described herein, the business continuity policy is implemented as a component of a cluster management application

FailOver Policy

One component of a business continuity policy is a FailOver Policy Several different FailOver Policies are possible, including Priority, Round Robin, and a Load FailOver policy included with the piesent invention

A Priority FailOver Policy is the most basic strategy The server system with the lowest priority in a running state is chosen as the failover target A "failover target" is a system selected to host the application groups that must be re-started For example, pπoπty can be set implicitly via ordering in a SystemList, such as SystemList = { server 1, server2} or explicitly by setting priority in the SystemList, such as SystemList = {systeml=0, system2=l } The Priority FailOver Policy strategy works well for a simple two-node cluster, or for a small cluster with a small number of application groups

A Round Robin FailOver Policy chooses the server system running the smallest number of application groups as a failover target Round Robin FailOver Policy is often used for larger clusters running a large number of application groups having essentially the same server load characteristics (for example, servers running similar databases or applications)

The Load FailOver Policy described herein enables a framework for server consolidation at the data center In a preferred embodiment, Load FailOver Policy takes into account System Capacity, Application Group Load, System Limits and Application Group Prerequisites

Load FailOver Policy Capacity and Load

In one embodiment, a system Capacity variable, also referred to herein as Capacity, for a system is set to a fixed value representing the system's load handling capacity An application group Load variable, also referred to herein as Load, for an application is set to a fixed demand (Load) placed on a processor by the application group For example, consider a 4-node cluster consisting of two 16- processoi servers and two 8-processor servers The administrator sets a Capacity value on the 16-CPU servers to 200 and the 8-CPU servers to 100 These Capacity values can be arbitrarily assigned but should reflect differences in capacity of the respective systems

Similarly, each application group running on a system has a predefined Load value When an application group is brought online, the application group's Load is subtracted from the available capacity of the system In one embodiment, a cluster management application keeps track of the available capacity of all systems in the cluster using an AvailableCapacity variable for each system AvailableCapacity is determined by subtracting Load of all applications groups online (an application group is considered online if the application group is fully or partially online) on a system from the system's Capacity When a failover occurs, the cluster management application determines the system with the highest AvailableCapacity and starts the application group on that system Duπng a failover scenario involving multiple application groups, failover decisions can be made seπally to facilitate the proper load-based choice, however, online operations to bring applications online on alternate systems can be performed in parallel

Capacity is a soft restriction, indicating that the value of AvailableCapacity can fall below zero Duπng a cascading failure scenario, AvailableCapacity can be negative

Load FailOver Policy Static Load vs Dynamic Load

The dynamic load of a server can be calculated using a formula AvailableCapacity ^■= Capacity - (Sum of Load of all online application groups) An alternative strategy for determining dynamic load is provided by some cluster management applications, including early versions of VERITAS Cluster Server (VCS) prior to VCS 2 0 These cluster management applications allow an administrator to determine a dynamic load of a server with an outside monitoring program and set a DynamicLoad variable to reflect the value determined The administrator can run any monitoπng package desired, and then provide an estimated load to the cluster management application If DynamicLoad is so provided, this value can be used to override calculated Load values, for example, AvailableCapacity can be calculated using the formula AvailableCapacity = Capacity - DynamicLoad This calculation allows an administrator to control system load more accurately than using estimated application group loading

However, the administrator must set up and maintain a load estimation package in addition to the cluster management application In some cluster management applications using a Load FailOver Policy, the system with the lowest value in the DynamicLoad variable is chosen for a failover target

In summary, available capacity of all systems to host application groups can be calculated using the following formula

AvailableCapacity of a system = Capacity - Cuπent System Load where

Cuirent System Load = Dynamic system load if dynamic system load variable is specified

OR

Sum of Load of all application groups online on the system Load FailOver Policy Limits and Prerequisites

System Limits and application group Prerequisites can also be used in a business continuity policy An administrator can provide the finite resources available on each system (Limits), such as shared memory segments, semaphores and other system resources For example, a particular server may be capable of hosting no more than two database applications Furthermore, a set of Prerequisites, each of which corresponds to available system resources and/or capacity, can be established for each application group For example, a particular database application may need, and have a Prerequisite indicating, five shared memory segments and twenty semaphores

In one embodiment, all of the Prerequisites specified in an application group's set of Prerequisites must be met before the application group can be started In one embodiment, system

Limits cannot be overridden, such that a system cannot be not chosen as a failover target if the system has already reached it's allowed Limits

Under the business continuity policy of the present invention, a set of eligible systems that meet the failed application group's Prerequisites, which can be equivalent to the application group's Load, is identified This set can be limited to only those systems that also have sufficient

AvailableCapacity to accept the failed application group and remain within the system's Limits From this set of eligible systems, the least loaded system can be selected as a failover target A system that does not meet all the Prerequisites of an application group cannot be selected as a failover target When a decision is made to bring an application group online on a particular system, the values of the set of Prerequisite variables for the system resources required for the application group are subtracted from the Current Limits of the system to indicate that these system resources are already allocated

In one embodiment of the invention, administrators first define application group Prerequisites and then define corresponding Limits for each system In this embodiment, each system can have different Limits, and only the Prerequisites and Limits applicable to each application group and system are required to be defined. If a system has no defined Limits for a given system resource, then a default value of 0 can be assumed. Similarly, when Prerequisites are not defined for a given system resource, a default value of 0 can be assumed

As an example of definitions of the Prerequisites and Limits vanables, the following configuration can be established to allow only one group online on a system at a given point in time Prerequisites = { GroupWeight = 1 }

Limits = { GroupWeight = 1 }

By specifying a Prerequisite GroupWeight value of one, only one application group can be online at a given time In addition, by specifying a Limits GroupWeight value of one for each system, each system can have only one application group online at a time The GroupWeight value can be consideied to represent the number of application groups that can be brought online When the GroupWeight value is zero, no more application groups can come online on that particular system For example consider a system having two systems, SI and S2. each specifying a Limit of GroupWeight = 1 The system also has three application groups, G 1 , G2 and G3 Groups G 1 and G2 have Prerequisites of GroupWeight = 1, and group G3 has no Prerequisites A Prerequisite of GroupWeight = 1 for Gl and G2 indicates that each of Gl and G2 requires one "unit" of GroupWeight to be brought online When G 1 goes online at S 1 , S l 's CurrentLimits become GroupWeight = 0, thus preventing G2 from also going online on SI G3, having no Prerequisites, can go online on either SI or S2

Prerequisites and Limits can be used to determine a set of eligible systems on which an application group can be started during failover or upon startup Once a set of eligible systems meeting the Prerequisites and Limits is identified, the established FailOver Policy dictates which of the set of eligible systems is selected as the failover target

Example System and Application Group Attributes

Table 1 below provides an example of one embodiment including system attπbutes that can be used to implement the business continuity policy of the present invention Table 2 provides examples of application group attributes

Table 1: System attributes

<img file="WO03102772A2_D0001.tif" /><img file="WO03102772A2_D0002.tif" />

Table 2: Application Gioup Attributes

<img file="WO03102772A2_D0003.tif" /><img file="WO03102772A2_D0004.tif" />

Establishing Application Group and System Configurations

The following configuration file, main cf, illustrates a system definition and an application group definition include "types cf ' cluster SGWM-demo ( ) system LargeSvrl (

Capacity = 200

Limits = { ShrMemSeg=20, Semaphores=100, Processors=12}

LoadWarningLevel = 90

LoadTimeThreshold = 600

)

group Gl (

SystemList = { LgSvrl, LgSvr2, MedSvrl, MedSvr2 }

SystemZones = { LgSvrl=0, LgSvτ2=0, MedSvrl=l, MedSvr2=l ] AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2 }

FailOverPohcy = Load

Load = 100

Prerequisites = { ShrMemSeg=10, Semaphores=50 Processors=6 } }

)

Using Capacity and Prerequisites

Using Capacity and Prerequisites together enables determination of a suitable failover system In one embodiment, the system meeting the Prerequisites for a given application group and having the highest AvailableCapacity is selected If multiple systems satisfy the Prerequisites for the given application gioup and have the same AvailableCapacity, the first system in the SystemList can be chosen Note that a system meeting the Prerequisites for an application group may not be eligible to host the application group if the system's Limits are already met The system's Limits are already met when the Current Limits for the system allow sufficient resources to meet the Prerequisites for the given application group

As mentioned earlier, in one embodiment, Capacity is a soft limit The system with the highest AvailableCapacity value can be selected, even if a negative AvailableCapacity value is produced when the application group is started on the system

Overload Warning

In one embodiment, an overload warning is provided as part of the Load FailOver Policy When a server sustains a pre-detei mined load level set by a LoadWarningLevel variable (statically or dynamically determined) for a predetermined time, set by a LoadTimeThreshold vaπable, an overload warning is initiated The overload warning can be provided by a usei -defined script or application designed to implement the FailOver Load Policy of a given enterprise For example, the user-defined script may provide a message on a console for the operator, or the user-defined script may move 01 shut down application groups based on usei-defined priority values For example, if Load on a server running a business critical database reaches and stays above a user-defined threshold, operators can be immediately notified The user-defined script could then scan the system foi any application groups with a lower priority than the database, such as an internal Human Resources application, and shut down or move the lower-pπoπty application to a system with a smaller current Load

System Zones

In one embodiment, SystemZones are used to designate a preferred subset of systems from which to select in an initial failover decision A cluster management application implementing a business continuity policy tries to re-start an application group within the application group's zone before choosing a system in another zone For example, consider a typical 3-tιer application infrastructure with web servers, application servers and database servers The application and database servers can be configured in a single cluster Using SystemZones enables the cluster management application for an application group to try to fail to another application zone server if another application zone server is available If another application zone server is not available, the cluster management application can try to failover to the database zone based on Load and Limits In this configuration, excess Capacity and Limits available in the database zone are reserved for the laiger load of a database failover, while application servers handle the Load of application groups in the application zone During a cascading failure, excess capacity in the cluster remains available to application groups The SystemZones feature allows fine tuning application failover decisions, yet retains the flexibility to failover anywhere in the cluster if necessary

Load-Based AutoStart

In one embodiment, the concepts of the Load FailOver Policy can also be used to determine where an application group should come up when the cluster initially starts Administrators can set an AutoStartPohcy variable to Load and allow the cluster management application to determine the best system on which to start the application group Application groups can be placed in an AutoStart queue for load-based staitup when the cluster management application determines the available systems As with failover, a subset of systems is first created that meet the Prerequisites and Limits, then of those systems, the system with the highest AvailableCapacity can be chosen

Using AutoStartPohcy = Load and SystemZones together allows the administrator to establish a list of preferred systems in a cluster to initially run an application gioup As mentioned above, in a 3- tiei architecture, the administrator can designate that application groups start first in the application zone and database groups start in the database zone Using Application Priorities in Conjunction with the Load FailOver Policy

By combining the Load FailOver Policy described above with application priorities, a truly automated business continuity policy for mission/business critical applications is provided This business continuity policy adds the necessary business intelligence to the cluster framework to make policy driven decisions at time of failure to best maintain critical applications and application performance

Application group Priorities allow the administrator to specify the relative importance of an application group over other application groups During any failure event, the cluster management application can determine a suitable failover system based on application group Priorities, Load and Limits For most single application group or single server failures, most clusters will have adequate spare capacity However, in a situation involving multiple failures, or reduced cluster capacity following a Disaster Recovery event, more difficult decisions may be required

Application group Priorities effectively provide a mechanism for the cluster to provide triage The most critical application groups remain functional, at adequate performance levels, at the possible expense of lower priority applications

In one embodiment, the following priorities can be assigned to an application group

Priority 1 - Mission Cntical

Prioπty 1 application groups must remain online and be restarted immediately upon failure The cluster management application can be configured to avoid stopping or moving Prioπty 1 application groups, unless the application group specifically faults or the operator intervenes Priority 1 application groups can sustain only the downtime necessary to restart the application group

Prioπty 2 - Business Critical

Priority 2 application groups are only slightly less important than Pπoi lty 1 application groups The cluster management application must keep these application groups online, but may perform a switchover, moving the Priority 2 application group to another servei to maintain cluster Load characteristics

Piioπty 3 - Task Critical

Pi loi lty 3 application groups may be moved at will to maintain cluster loading Priority 3 application groups also may be stopped to maintain adequate Load handling capability in the cluster, but only if a move is not possible Pπoπty 4 - Task Non-Critical

Pi loi lty 4 Application gioups are non-essential applications such as test applications or various internal support programs These application groups may be stopped at will to maintain cluster loading During any cluster reconfiguration, the cluster management application can remove all Priority 4 application groups from the calculation and make its best recommendation for reconfiguration Priority 4 applications may only be brought online m the cluster if the cluster management application determines that there is adequate load capacity remaining in the cluster

Fig 3 is a flowchart of a method for implementing a business continuity policy in a server consolidation environment The method begins upon startup or failure of a given application group, here called application group X, in Startup or Failure of Application Group X step 310 A set of systems eligible to host application group X is identified in Determine Set of Eligible Systems to Host Application Group X step 320 At Size of Set &gt; 0 decision point 322, a determination is made whether any eligible systems were identified If so, control proceeds to Select Host System 324 to select a host system (either an initial system upon startup or a failover target) for running application group X For example, the host system can be selected as the eligible system having the highest Available Capacity Other policies can also be used to select a host system according to the needs of the business implementing a business continuity policy Control then proceeds to Start Application Group X on Host System step 350 to start application group X on the selected host system

If at Size of Set &gt; 0 decision point 322, the set includes no eligible systems for hosting application group X, control proceeds to Determine Priority of Application Group X step 330 A respective prioπty for application group X among all application groups running on the cluster is determined The priority of a given application group is configurable and can be assigned by an administratoi of the server consolidation environment For example, to determine the respective priority for application group X, the pπoπty can be retneved from data stored for a cluster management application managing the cluster in the server consolidation environment

From Determine Priority of Application Group X step 330, control proceeds to Lower Priority Application Groups in Cluster decision point 332 If no lower priority applications are running, control proceeds to Notify Administrator that Application Group X Cannot be Started step 336 Because no eligible systems exist for application group X, application group X cannot be started without pre- empting another application of the same or higher priority An administrator can determine whether

Application Group X should be pie-empted In one embodiment, the process for handling the situation where an application group cannot be restarted is configurable within a cluster management application and can be provided as a user-defined script If at Lower Prioπty Application Groups in Cluster decision point 332, lower priority application groups are running, control proceeds to Can Sufficient Capacity and Resources be Freed to Accommodate Application Group X decision point 338 In Can Sufficient Capacity and Resources be Freed to Accommodate Application Group X decision point 338, an evaluation of the available resources in the systems of the cluster is made This evaluation is discussed in further detail with reference to Fig 5

If sufficient capacity and resources cannot be freed, control proceeds to Notify Administrator that Application Group X Cannot be Started step 336 If sufficient capacity and resources can be freed, control proceeds to Free Sufficient Capacity and Resources on Host System step 340

In Free Sufficient Capacity and Resources on Host System step 340, capacity and resources are freed on one or more systems to enable sufficient resources for application group X to run on a given host system From Free Sufficient Capacity and Resources on Host System step 340, control proceeds to Start Application Group X on Host System step 350

Fig 4 is a flowchart of the Determine Set of Eligible Systems to Host Application Group X step 320 of Fig 3 In Select System from Cluster step 410, a system within the cluster of systems not previously evaluated is selected to determine whether the system is eligible Control then proceeds to Selected System Meets Application Requirements decision point 412 If the selected system does not meet the requirements for application group X, such as a prerequisite for application group X, control proceeds to Systems Not Considered Remam decision point 422 to determine whether another system is available for evaluation

If the selected system meets the requirements for application group X, control proceeds to Selected System Meets System Requirements decision point 414 For example, a determination whethei the selected system is within its Limits can be made by adding the system's Current Limits to the Pierequisites for Application Group X The sum must be less than the Limits of the Selected System to meet the Limits criterion As another example, a system requirement may be that a particular CPU remains below a certain utilization percentage If the selected system does not meet the system requirements, control proceeds to Systems Not Consideied Remain decision point 422 to determine whether another system is available for evaluation

If the selected system meets the system requirements at Selected System Meets System Requirements decision point 414, control proceeds to Add Selected System to Set of Eligible Systems step 420 Control then proceeds to Systems Not Considered Remain decision point 422 to determine whethei another system is available for evaluation

In Systems Not Considered Remain decision point 422, a determination is made whether any systems not already considered remain in the cluster If so, control proceeds to Select System step 410 to select another system If not, the set of eligible systems is complete and control returns to Size of Set &gt; 0 decision point 322 of Fig 3

Fig 5 is a flowchart of the Can Sufficient Capacity and Resources be Freed to Accommodate Application Group X decision point 338 of Fig 3 The initial decision is made at Can Sufficient Prioπty 4 Resources be Stopped decision point 510 If sufficient Priority 4 resources can be stopped, control proceeds to Select Host System and Priority 4 Resources to Free step 520 In this step, a system with sufficient Prioπty 4 resources is selected as the system to host application group X Control proceeds to Indicate Sufficient Resources can be Freed step 565 The flowchart of Fig 5 is completed and an indication that sufficient resources can be freed is made

If at Can Sufficient Pπoπty 4 Resources be Stopped decision point 510, sufficient Prioπty 4 resources cannot be freed, control proceeds to Can Sufficient Pπoπty 4 Resources be Stopped and Priority 3 Resources Moved decision point 530 If priority 4 applications can be stopped and sufficient resources for Application Group X freed on a system by moving priority 3 applications to other systems, then control proceeds to Determine Appropriate Prioπty 3 and 4 Resources to Free and Select Host System step 540 At Determine Appropriate Pπoπty 3 and 4 Resources to Free and Select Host System step 540, the decision of which priority 4 applications to stop and which prioπty 3 applications to move is made Preferably, when several different scenarios can free the necessary resources, a configuration can be selected such that a minimum number of resources are stopped and or moved to enable the largest number of high-priority applications to run Control then proceeds to Indicate Sufficient Resources can be Freed step 565 The flowchart of Fig 5 is completed and an indication that sufficient resources can be freed is made

If at Can Sufficient Priority 4 Resources be Stopped and Priority 3 Resources Moved decision point 530, sufficient resources are not available, control proceeds to Can Sufficient Priority 4 Resources be Stopped and Priority 2 and 3 Resources Moved decision point 550 If so, control proceeds to Determine Appropriate Priority 2, 3 and 4 Resources to Free and Select Host System step 560 Again, preferably minimal resources are stopped and moved to enable the largest number of high-pπoπty applications to run Control then proceeds to Indicate Sufficient Resources can be Freed step 565 The flowchart of Fig 5 is completed and indication that sufficient lesources can be freed is made

If at Determine Appropriate Pπoπty 2, 3 and 4 Resources to be Freed and Select Host System step 560, sufficient resources are not available in the cluster, control proceeds to Indicate Sufficient Resouices Cannot be Freed step 570 The flowchart of Fig 5 is completed and an indication that sufficient resources cannot be freed is made

Tigs 6 through 16 describe multiple scenarios that are within the scope of the business continuity policy of the present invention Fig 6 shows the calculation of available capacity for a cluster of servers in a server consolidation environment Servers 610A, 61 OB, 6 IOC and 610D form a cluster Servers 610A, 61 OB and 6 IOC each have a capacity of 300, and server 610D has a capacity of 1 0 Server 610A is running Microsoft Exchange (XCH) version 5 5, which places a Load of 100 on server 610A Server 610A is also running a database application group, Oracle 8ι, which places a Load of 150 on server 610A, for a total Load of 250 Server 610B is running SQL 2000 server, which places a Load of 125 on server 610B Server 610C is running a file sharing application group FileSharel , which places a Load of 75 on Server 610C Server 610D is running two file sharing application groups, Fιleshare2 and Fιleshare3, placing a load of 150 on server 610D By subtracting the respective Load for each application group running on a given server from the Capacity of the given server, Available Capacity is calculated as 50 for server 610A, 175 for server 610B, 225 for server 610C, and zero for server 610D With an available capacity of 225, server 610C has the highest available capacity in the cluster

Fig 7 shows the movement of an application upon failure of one of the servers of Fig 6 and the resulting available capacity in the cluster Server 610D fails, leaving file sharing applications Filesharel and Fιleshare2 to be redistributed if possible to other servers in the cluster Fig 7 shows the movement of Fιleshare2 to server 610C, which is selected because server 610C offers the highest available capacity As a lesult of the movement of Fιleshare2 to server 610C, the Load on server 610C increases to 150, and the available capacity of server 6 IOC reduces to 150 Server 610B, with an available capacity of 175, now has the highest available capacity in the cluster

Fig 8 shows the movement of another application in the failure scenario of Fig 7 Fιleshare3 is moved from server 610D to the server having the highest available capacity, server 610B As a result of this move, the Load placed on server 610B is increased to 200 and the available capacity of server 610B is reduced to 100

Fig 9 shows an example configuration of database applications in the cluster of Fig 6, with each of servers 610A thiough 610D configured with a capacity of 300 Server 610A is running two SQL 2000 database application groups, SQL 2000 Database A and SQL 2000 Database B Each of SQL 2000 Database A and SQL 2000 Database B places a load of 100 on server 610A Server 610A is configured with an SQL limit of 2, indicating that server 610A can run no more than two SQL databases at one time The available capacity on servei 610A is 300 - 200 = 100

Server 610B similarly has a SQL limit of 2 and is running SQL 2000 Database C, placing a load of 100 on server 610B Server 610B has an available capacity of 200 Server 610C is running SQL 2000 Database E, placing a load of 100 on server 610C Server 610C also has an available capacity of 200 Server 610D has a SQL limit of 3 and is running SQL 2000 Database D, which places a Load of 150 on server 610D Server 610D has an available capacity of 150 Fig 10 shows movement of database applications in a failure scenario m the configuration of Fig 9 Seiver όlOC fails, leaving SQL 2000 Database E to be lestarted on another server SQL 2000 Database E places a Load of 100 on a seiver Server 610A cannot host SQL 2000 Database E because sever 610A has already reached its limit of 2 server SQL applications Neither server 610B or server 610D has reached its limit of the number of SQL applications that it can host, and both server 610B and server 610D have sufficient available capacity to run SQL 2000 Database E In the example scenaπo shown, server 610B is selected because, of the two eligible systems, server 610B has the highest available capacity After SQL 2000 Database E is moved, the load placed on server 610B increases to 200 and the available capacity of server 610B reduces to 100

Fig 11 shows an example of managing application groups using limits and prerequisites

Four application groups are given in this example, including application group Gl, a file sharing application, application group G2, a test application, application group G3, a Microsoft Exchange application, and application group G4, a SQL server application group Application group Gl, a pnoπty three application group, requires that a GroupWeight vaπable for the server have a value of 1 before application group Gl can be run on that server Application group G2, a priority four application group, requires that a GroupWeight vaπable for the server have a value of 2 before application group G2 can be run on that server Application group G3, a prioπty one application group, requires that a GroupWeight variable for the server have a value of 2 before application group G3 can be run on that server Finally, application group G4, a pπoπty two application group, requires that a GroupWeight variable for the server have a value of 2 before application group G4 can be run on that server

Servers 610A through 610D run applications Gl through G4, respectively With these running application groups, servers 610A through 610D have Limits of 2, 3, 2 and 3, respectively Servei s 610A through 610D have CurrentLimits values of 1, 1, 0, and 1, respectively

Fig 12 shows a failure scenario in which an application group cannot be failed over Server

610C fails, and no server has a CurrentLimits value of 2, which is a prerequisite for application group G3 to be started on another server When an application group cannot be failed over, pπoπties of the running applications are examined to determine whether sufficient resources can be freed in the cluster to run the application group Application group G3 is a prioπty one application, and each of application groups G2 tlirough G4 is a lower prioπty application group Fust, a determination is made whether sufficient priority 4 resources exist to free sufficient resources for application group G3 Application group G2 is a priority 4 resouice, and it consumes two GroupWeight units If application group G2 is freed, the two GroupWeight units necessary to run application group G3 are freed, and application group G3 can be started on server 610B

Fig 13 shows stopping a lowei -prioπty application group to free sufficient resources to enable a higher-pi loπty application group to remain available In the scenario of Fig 12, application group G2 was determined to provide sufficient resources to allow application group G3 to run Application group G2 is stopped, and application group G3 is moved to servei 61 OB The CurrentLimits value for server 61 OB is recalculated, now having a value of 1

Fig 14 shows another failure scenario for the configuration of Figs 12 and 13 Assume that now server 610D fails, leaving application G4 to be restarted Application group G4 requires a GroupWeight value of 2 to be started on another server Neither of the remaining servers 610A or 61 OB provides a GroupWeight value of 2 A determination is then made whether sufficient resources can be freed to enable application group G4 to remain available Lower priority resources are examined to make this determination

Fig 15 shows movement of a lower-priority application group to free sufficient resources to enable a higher-priority application group to remain available Application group Gl, a priority three application, has a lower pnoπty than application group G4, with a prioπty of two Furthermore, by moving application group Gl, the value of GroupWeight for server 610A can be raised to two, which meets the prerequisite for application group G4 The prerequisites for application group G 1 are a GroupWeight value of 1, which is provided by server 61 OB Application group Gl is moved to server 61 OB to free resources on server 610A The result of the movement is that server 610A has a GroupWeight value of 2, and server 61 OB has a GroupWeight value of zero

Fig 16 shows movement of the higher-priority application group to use the resources freed as a result of the action shown in Fig 15 After the movement of application group Gl, server 610A has sufficient resources to host application group G4 The pierequisite for application group G4, that

GroupWeight have a value of 2, is true After the movement of application group G4, server 610A has a GroupWeight value of zero

The above scenarios are examples of multiple failure situations that can be handled by the business continuity policy described herein Many variations of these scenarios, and alternative variables for implementing the business continuity policy, are envisioned as part of the present invention and fall within its scope Further example scenarios are provided in the "Additional Examples" section of this document

Resource Manager Integration

Most major operating systems have a corresponding resource manager, such as Solans resource manager, HP Process Resource Manager and A1X Resource manager These resource manageis, collectively called xRM here, allow an administrator to control CPU and memory utilization However, typically xRM packages are only aware of the system on which the xRM package is running, and not of other systems within the cluster Preferably, a cluster management application supporting the business continuity policy of the present invention is integrated with <img file="WO03102772A2_D0005.tif" /> packages and controls lesource utilization, and therefore Load, on all systems in the cluster

Each operating system vendor provides a different interface and different capabilities in the platfoim's resource manager For example, Solans 9 supports the concept of a "Task-ID," which ties a specific process launched under the Task-ID to limits imposed in a "projects" database To maintain maximum flexibility and keep operations across the operating system platforms identical, a cluster management application provides an API layer to communicate with the various xRM packages At a minimum, the Load FailOver policy can be used If the cluster management application is also running on an operating system platform capable of xRM integration, then full enforcement of Load and Limits is possible

In one embodiment, administrators can configure resource utilization parameters once m the cluster definition, rather than on individual systems The cluster management application, m conjunction with xRM-specific agents on each system, controls resource allocation to specific application groups when the application groups are started on a system This allows a single point of administration as well as greater control of load distπbution in the cluster

By changing values for application group Load, the administrator sets both the overall load an application group is expected to place on a system, as well as the share of a system the application group is expected to receive For example, if three application groups with a Load of 200 each were running on a server with a capacity of 800, each application group effectively receives 1/3 of the available resources In this scenario, raising the Load value for a specific application group to 400 accomplishes several things First, raising the load value increases the resource allocation for the modified application group This application group receives 50% of available CPU and memory, and the remaining two application groups receive 25% each Second, raising the Load Value places the server at a 100% load level, reducing AvailableCapacity to 0 This situation produces an overload warning Raising a Load value not only tells the cluster management application that a system is loaded more heavily, it also functions to increase the performance of the application

Modeling and Simulation Engine

A modeling and simulation engine (MSE) can provide the capability for the cluster management application to determine the best possible configuration for application groups based on a "what-if ' model Rather than choose a system based solely on current load and limits, the cluster management application determines how to reconfigure the cluster to provide application groups with the best possible performance Re-configuration takes into account the vanous application group pπoi ities to determine the application groups that can and cannot be moved Various parameters can also be supplied to the MSE, such as "maximum performance" and "minimum switches," to allow the cluster management application to deteimine whether to perform a cluster reconfiguration to maximize performance or whether to minimize downtime for application groups

The MSE can also include simulation capabilities to allow administrators to run a complete what-if scenario for any cluster reconfiguration For example

What if I take 32 CPU server- 1 out of the cluster⁷ What is the best performance reconfiguration model⁹ What applications will be stopped due to the shutdown⁹ What applications will be stopped due to reconfiguration moves⁹ What if I allow Prioπty 1 moves during this evolution⁹

What if I add an additional four 16-CPU commodity servers to my cluster and storage area network ^? What is the best performance configuration⁹ What applications will be stopped duπng the move⁹ How much spare capacity will this configuration provide⁹

I want to online a large database Where is the best location⁹ What reconfiguration would provide the best fit

The MSE can ngidly enforce the current concepts of Load and Limits, and also allows reconfiguration to better utilize the FailOver Policy For example, to add a large database (shared memory and semaphores X2) and no system has adequate capacity within the Limits, the proposed FailOver Policy provides an error The MSE could determine that two systems piovide available adequate resources, but each is running a small database (shared memory and semaphores The cluster management application can recommend a consolidation of the two smaller databases to one server and free the second server for the large database

Cluster Reconfiguration

Clustei Reconfiguration, either manual or automatic, refers to the capability piovided by the clustei management application to re-allocate application groups, and therefore loads across the cluster, to better balance system Load This re-configuration can be in response to a failuie, server additions and deletions, or application group additions or removals Cluster reconfiguration can be performed by an MSE component of the cluster management application to allocate fixed cluster resouices The cluster reconfiguration module can be allowed to perform automatically if moving Pi loπty 3 and Priority 4 application gioups, and possibly automatically on Priority 2 application groups if specific parameteis are set and manual (operator-acknowledged) for Priority 1 groups

Clustei reconfiguration capabilities can intervene when a manual application group online or switchovei is requested If a user requests to move or bring an application group online, the MSE can inform the usei that it is acceptable or recommend a reconfiguration sequence to better allocate resources Additional Examples

The following example uses Limits and Preiequisites to control the total number of application groups that may run on a system The cluster consists of four similar servers There are five application groups, which are roughly equivalent in requirements for processing power and in the amount of Load each application group requires of a system Each server can host two such application groups This example does not use application group Load and system Capacity Also, the application groups use a default AutoStartPohcy and FailOverPohcy

Example Configuration File with Limits

system Svrl (

Limits = {GroupWeight = 2} ) system Svr2 (

Limits = {GroupWeight = 2} ) system Svr3 (

Limits = {GroupWeight = 2}

) system Svr4 ( Limits = {Gioup Weight = 2}

) group G 1 (

SystemList = { Svrl, Svr2, Svr3, Srv4) AutoStartList = { Svrl, Svr2 } Prerequisites = { GroupWeight = 1 }

) group G2 (

SystemList = { Svrl, Svr2, Svr3, Svt4} AutoStartList = { Svr2, Svr3 } Prerequisites = { Gioup Weight = 1 }

) group G3 (

SystemList = { Svrl , Svr2, Svr3, Svr4} AutoStartList = {Svr3, Svr4 } Preiequisites = { GroupWeight = 1 }

) group G4 (

SystemList = { Svrl , Svr2, Svr3, Svr4} AutoStartList = { Svr4, Svrl J Prerequisites = { GroupWeight = 1 }

) group G5 (

SystemList = { Svrl, Svr2, Svr3, Svr4} AutoStartList = { Svr2, Svr3 } Prerequisites = { GroupWeight = 1 | )

AutoStai t Operation

This example uses the default AutoStartPohcy = Ordei Application groups are brought online on the first system available in the AutoStartList In this way, Gl will start on Svrl , G2 on Svr2, and so on G5 will start on Svr2

Normal Opei ation

An example cluster configuration (assuming all systems are running) is provided below

Svrl

CurrentLimits = { Group Weιght=l ] (Group Gl)

Svr2

CurrentLimits = {GroupWeight=0} (Groups G2 and G5)

Svr3

CurrentLimits = {GroupWeιght=l j (Group G3)

Svr4

CurrentLimits = {GroupWeιght=l } (Group G4)

Faύui e Scenaπo

In the first failure scenaπo, assume Svr2 tails With application groups G2 and G5 configured with an identical SystemList, both application groups can run on any system The cluster management application can serialize the choice of failovei nodes for the two groups G2, being canomcally first, is started on Svrl, the lowest priority in the SystemList, thereby exhausting the Limits for Svrl G5 is then started on the next system in the order of the SystemList for group G5 G5 goes online on Svr3 Following the first failure, the cluster now looks like the following

Svrl

CurrentLimits = { Group Weιght=0 (Groups Gl and G2) Svr3

CurrentLimits = {GroupWeight=0} (Groups G3 and G5)

Svr4 CurrentLimits = {Group Weιght=l }

(Gioup G4)

Cascading Failures

Assuming Svr2 cannot immediately repaired, the cluster can tolerate the failure of an individual application group on Svrl or Svr3, but no further node failures

Load-Based example

The following sample cluster shows the use of simple load based startup and failover SystemZones, Limits and Prerequisites are not used

The cluster consists of four identical systems, each with the same capacity Eight application groups, G1-G8, with various loads run in the cluster

Example Configuration File include "types cf ' cluster SGWM-demo system Svrl (

Capacity = 100 ) system Svr2 (

Capacity = 100 ) system Svr3 ( Capacity = 100

) system Svr4 (

Capacity = 100 ) group Gl (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPohcy = Load AutoStartList = { Svrl, Svr2, Svr3, Svr4 } FailOverPohcy = Load Load = 20

) group G2 (

SystemList = { Svrl , Svr2, Svι4. Svr4 | AutoStartPohcy = Load AutoStartList = { Svrl, Svr2, Svr3, Svr4 1 FailOverPohcy = Load

Load = 40 ) group G3 (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPohcy = Load

AutoStartList = { Svrl, Svr2, Svr3, Svr4 } FailOverPohcy = Load Load = 30 ) group G4 (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPohcy = Load AutoStartList = { Svrl, Svr2, Svr3, Svr4 } FailOverPohcy = Load Load = 10

) group G5 (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPohcy = Load AutoStartList = { Svrl, Svr2, Svr3, Svr4 }

FailOverPohcy = Load Load = 50 ) group G6 (

SystemList = { Svrl, Svr2, Svr4, Svr4 }

AutoStartPohcy = Load

AutoStartList = { Svrl, Svr2, Svr3, Svι4 }

FailOverPohcy = Load

Load = 30 ) group G7 (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPohcy = Load AutoStartList = { Svrl, Svr2, Svr3, Svr4 } FailOverPohcy = Load

Load = 20 ) group G8 (

SystemList = { Svrl, Svr2, Svr4, Svr4 } AutoStartPolicy = Load AutoStartList = { Svrl , Svr2, Svr3, Svr4 } FailOverPohcy = Load

Load = 40 )

AutoStart Operation

As mentioned above, application groups can be placed in a queue as soon as they are started on a system For the purposes of this example, application groups are placed into the queue in the same order that the application groups are described, Gl through G8

Gl is started on the system with the highest AvailableCapacity Since the systems are equal, Svrl is chosen smce it is canomcally first G2-G4 start on Svr2 through Svr4 At this time, with the first 4 group startup decisions made, the cluster looks as follows

Svrl

AvaιlableCapacιty=80

Svr2 A vailableCapac ιty=60

Svr3 AvaιlableCapacιty=70

Svr4

AvaιlableCapacιty=90

As the remaining application groups are brought online, G5 starts on Svr4, as it has the highest AvailableCapacity G6 are brought starts on Svi 1, with 80 remaining G7 starts on Svr3, with AvaιlableCapacιty=70 G8 starts on Svr2, with AvaιlableCapacιty=60

Normal Opei atwn

The final cluster configuration (assuming the original queue of G1-G8) is shown below

Svrl

AvaιlableCapacιty=50 (Groups Gl and G6)

Svr2

AvaιlableCapacιty=20 (Groups G2 and G8) Svr3

AvaιlableCapacιty=50 (Groups G3 and G7)

Svr4 AvaιlableCapacιty=40

(Groups G4 and G5)

In this configuration, an overload warning is provided for Svr2 after the default 900 seconds since Svr2 has a default LoadWarningLevel of 80%

Faihtr e Scenario

In the first failure scenario, assume Svr4 fails, immediately queuing G4 and G5 for failure decision G4 starts on Svrl, as Svrl and Svr3 have AvaιlableCapacιty=50 and Svrl is cano cally first G5 goes online on Svr3 Svrl Failure decisions are made serially, actual online and offline operations are not Senahzing the failover choice allows complete load-based control, and, in one embodiment, adds less than one second to total failover time

Following the first failure, the cluster configuration is shown below

Svrl

AvaιlableCapacιty=40 (Groups Gl, G6 and G4)

Svr2 AvaιlableCapacιty=20

(Groups G2 and G8)

Svr3

AvailableCapacity=0 (Groups G3, G7 and G5) In this configuration, an overload warning is provided for Svr3 to notify an operator or administrator that Svr3 is overloaded The opeiator can switch G7 to Svrl to balance loading across Gl and G3 As soon as Svr4 is repaired, Svr4 rejoins the cluster with an AvaιlableCapacιty=100 Svι4 can then server as a failovei target for further failures

Cascading Failui es

Assuming Svr4 is not immediately repaired, further failures are possible For this example, assume Svr3 now fails Each application group G3 G5 and G7 is re-started on respective server Svrl , Svr2, and Svrl These re-starts result in the following configuration <img file="WO03102772A2_D0006.tif" />

AvaιlableCapacιty= -10 (Groups G 1 , G6, G4, G3 and G7)

Svr2 AvaιlableCapacιty= -30

(Groups G2 and G8 and G5)

This example shows that AvailableCapacity is a soft limit, and can fall below zero

Complex 4-system Example

The following example shows a 4-system cluster using multiple system Capacities and various Limits The cluster consists of two large Enterprise servers (LgSvrl and LgSvr2) and two Medium servers (MedSvrl and MedSvr2) Four application groups, Gl through G4, are provided with various Loads and Prerequisites Gl and G2 are database application groups, with specific shared memory and semaphore requirements G3 and G4 are middle-tier application groups with no specific memory or semaphore requirements and simply add load to a given system

Example Configuration File

include "types cf ' cluster Demo ( ) system LgSvrl (

Capacity = 200

Limits = { ShrMemSeg=20, Semaphores= 100, Processors=12} LoadWarningLevel = 90 LoadTimeThreshold = 600

) system LgSvr2 (

Capacity = 200

Limits = { ShrMemSeg=20, Semaphores= 100, Processors=12 } LoadWarnιngLevel=70 LoadTιmeThreshold=300 ) system MedSvrl (

Capacity = 100

Limits = { ShrMemSeg=10, Semaphores=50, Processors=6}

) system MedSvr2 (

Capacity = 100

Limits = { ShrMemSeg=10, Semaphores=50, Processors=6 }

) gioup Gl (

SystemList = j LgSvi l, LgSvr2, MedSvrl , MedSvr2 j SystemZones = { LgSvrl =0, LgSvr2=0, MedSvrl = l , MedSvr2=l } AutoStartPohcy = Load AutoStartList = { LgSvrl , LgSvr2 }

FailOverPohcy = Load Load = 100

Prerequisites = { ShrMemSeg= 10, Semaphores=50, Processors=6 } ) gioup G2 (

SystemList = { LgSvrl, LgSvr2, MedSvrl, MedSvr2 } SystemZones = { LgSvrl=0, LgSvr2=0, MedSvrl=l, MedSvr2=l } AutoStartPohcy = Load AutoStartList = { LgSvrl, LgSvr2 } FailOverPohcy = Load

Load = 100

Prerequisites = { ShrMemSeg=10, Semaphores=50, Processors=6 } ) group G3 ( SystemList = { LgSvrl, LgSvr2, MedSvrl, MedSvr2 }

SystemZones = { LgSvrl=0, LgSvr2=0, MedSvrl = 1, MedSvr2=l }

AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2 }

FailOverPohcy = Load Load = 30

) gioup G4 (

SystemList = { LgSvrl, LgSvr2, MedSvrl, MedSvr2 } SystemZones = { LgSvrl=0, LgSvr2=0, MedSvrl =1, MedSvr2=l } AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2 } FailOverPohcy = Load Load = 20 ) AutoStart Operation

Using the main cf example above, the following is one possible outcome of the AutoStart operation

Gl - LgSvrl G2 - LgSvr2 G3 - MedSvrl

G4 - MedSvr2

All application groups are assigned to a system when the cluster starts Application groups Gl and G2 have an AutoStartList of LgSvrl and LgSvr2 Gl and G2 are queued to go online on one of these seivers, based on highest AvailableCapacity Assuming Gl starts fust, Gl is started on LgSvrl because LgSvrl and LgSvι2 both have an initial AvailableCapacity of 200, and LgSvrl is lexically first

Application groups G3 and G4 are started, respectively, on MedSvrl and MedSvr2

Normal Opei ation

After starting application groups Gl through G4, the resulting configuration is shown below

LgSvrl

AvaιlableCapacιty=100

CurrentLimits = { ShrMemSeg=10, Semaphoi es=50, Processors=6 }

LgSvr2 AvatlableCapacιty=100

CurrentLimits = { ShrMemSeg=10, Semaphores=50, Processors=6}

MedSvrl

AvaιlableCapacιty=70

CurrentLimits = { ShrMemSeg=10, Semaphores=50, Processors=6 } MedSvr2

AvaιlableCapacιty=80

CurrentLimits = { ShrMemSeg=10, Semaphores=50, Pιocessoιs=6}

Failui e Scenai ω

For the first failure example, assume system LgSvr2 fails The cluster management application scans available systems in G2's SystemList having the same SystemZones grouping as

LgSvr2 The cluster management application then creates a subset of systems meeting the application group's Prerequisites In this case, LgSvrl meets all necessary Limits G2 is brought online on LgSvrl, resulting in the following configuration

LgSvrl AvailableCapacity=0

CurrentLimits = { ShrMemSeg=0, Semaphoi es=0, Processors=0 }

MedSvrl

AvaιlableCapacιty=70

CuirenfLimits = { ShrMemSeg=10, Semaphores=50, Processors=6 } MedSvr2

AvaιlableCapacιty=80

CurrentLimits = { ShrMemSeg=10, Semaphores=50, Processors=6}

After 10 minutes, (LoadTimeThreshold = 600) the oveiload warning on LgSvrl is provided because LoadWarningLevel exceeds 90% Ca cadmg Failure Scenai 10

In this scenario, a further failure of a system can be toleiated, as each system has remaining Limits sufficient to accommodate the application group running on the peer system

For example, if a failure were to occur with either MedSvrl or MedSvr2, the other system would be selected as a failover target, as application groups running on the failed system have MedSvrl and MedSvr2 in their respective SystemZones

If a failure instead occmred with LgSvrl, with LgSvr2 still offline, the failover of the application groups Gl and G2 are serialized for the failover decision process In this case, no systems exist in the database zone The first group canomcally, G 1 , will be started on MedSvr2, as MedSvr2 meets all Limits and has the highest AvailableCapacity Group G2 will be started on MedSvrl, as MedSvrl is the only remaining system meeting the Limits

Server Consolidation Example

The following example shows a complex 8-node cluster running multiple applications and several large databases The database servers are all large enterprise systems, LgSvrl, LgSvr2 and LgSvr3 The middle-tier servers running multiple applications are MedSvi 1 , MedSvr2, MedSvr3, MedSvr4 and MedSvr5

Example Configuration File include "types cf ' cluster Demo ( ) system LgSvrl (

Capacity = 200

Limits = { SlιrMemSeg=15, Semaphores=30, Processors=18} LoadWarningLevel = 80 LoadTimeThreshold = 900

) system LgSvr2 (

Capacity = 200

Limits = { ShrMemSeg=15, Semaphores=30, Processors=18 } LoadWaιnιngLevel=80

LoadTιmeThreshold=900 ) system LgS vi 3 (

Capacity = 200 Limits = { ShrMemSeg=15, Semaphores=30, Processors=18 }

LoadWarnιngLevel=80 LoadTιmeThreshold=900 ) system MedSvrl (

Capacity = 100

Limits = j ShrMemSeg=5, Semaphoi es= 10, Processoιs=6 j ) system MedSvr2 (

Capacity = 100

Limits = { ShrMemSeg=5, Semaphores= 10, Processors=6

) system MedSvr3 ( Capacity = 100

Limits = { ShrMemSeg=5, Semaphores= 10, Processors=6 } ) system MedSvr4 (

Capacity = 100 Limits = { ShrMemSeg=5, Semaphores=10, Processors=6 }

) system MedSvr5 (

Capacity = 100

Limits = { ShrMemSeg=5, Semaphores=10, Processors=6 } ) group Database 1 (

SystemList = { LgSvrl , LgSvr2, LgSvr3, MedSvrl, MedSvr2, MedSvr3,

MedSvr4, MedSvr5 } SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl=l, MedSvr2=l, MedSvr3=l, MedSvr4=l, MedSvr5=l }

AutoStartPohcy = Load AutoStartList = { LgSvrl, LgSvr2, LgSvr3 } FailOverPohcy = Load Load = 100 Prerequisites = { ShrMemSeg=5, Semaphores=10, Processors=6 }

}

) group Database2 (

SystemList = { LgSvrl, LgSvr2, LgSvι3, MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvι5 }

SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl = l, MedSvr2=l,

MedSvr3=l, MedSvι4=l, MedSvr5=l } AutoStartPohcy = Load AutoStartList = { LgSvrl, LgSvr2, LgSvι3 } FailOverPohcy = Load

Load = 100

Preiequisites = { ShrMemSeg=5, Semaphoi es= 10, Processors=6 } }

) jroup Database3 (

SystemList = { LgSvrl, LgSvr2, LgSvr3, MedSvrl . MedSvr2, MedSvr3,

MedSvr4, MedSvi 5 } SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl = l , MedSvr2=l,

MedSvr3=l, MedSvr4=l , MedSvr5=

AutoStartPohcy = Load AutoStartList = { LgSvrl, LgSvr2, LgSvr3 } FailOverPohcy = Load Load = 100 Prerequisites = { ShrMemSeg=5, Semaphores=10, Processors=6 }

}

) group Application 1 (

SystemList = { LgSvrl, LgSvr2, LgSvr3, MedSvrl , MedSvr2, MedSvr3, MedSvr4, MedSvr5 }

SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl=l , MedSvr2=l,

MedSvr3=l, MedSvr4=l, MedSvr5=l { AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2, MedSvr3, MedSvrf, MedSvr5 } FailOverPohcy = Load

Load = 50 ) group Application (

SystemList = { LgSvrl , LgSvr2, LgSvr3, MedSvrl , MedSvr2, MedSvr3, MedSvr4, MedSvr5 }

SystemZones = { LgSvrl=0, LgSvι2=0, LgSvr3=0, MedSvrl=l, MedSvr2=l,

MedSvr3=l, MedSvι4=l, MedSvr5=l } AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvr5 } FailOverPohcy = Load

Load = 50 ) group Application (

SystemList = { LgSvrl, LgSvr2, LgSvr3, MedSvrl , MedSvr2, MedSvr3, MedSvr4, MedSvr5 }

SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl=l, MedSvr2=l,

MedSvr3=l, MedSvr4=l, MedSvr5=l } AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvr5 } FailOverPohcy = Load

Load = 50 ) group Apphcatιon4 (

SystemList = { LgSvrl, LgSvr2, LgSvr3, MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvr5 }

SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl=l, MedSvr2=l,

MedSvr3=l, MedSvr4=l, MedSvr5=l } AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvr5 } FailOverPohcy = Load

Load = 50 ) group Apphcatιon5 (

SystemList = { LgSvrl, LgSvr2, LgSvr3, MedSvrl , MedSvr2, MedSvr3,

MedSvr4, MedSvr5 } SystemZones = { LgSvrl=0, LgSvr2=0, LgSvr3=0, MedSvrl = l , MedSvr2=l, MedS vr3= 1 , MedSvr4= 1 , MedSvr5= 1 J

AutoStartPohcy = Load

AutoStartList = { MedSvrl, MedSvr2, MedSvr3, MedSvr4, MedSvr5 } FailOverPohcy = Load Load = 50 )

AutoStart Opetation

Using the example configuration file above, the following AutoStart Sequence is possible

Database 1 - LgSvrl

Database2 - LgSvr2 Database3 - LgSvr3

Apphcationl - MedSvrl

Application - MedSvr2

Application - MedSvr3

Apphcatιon4 - MedSvr4 Apphcatιon5 - MedSvi 5

Normal Opeuilion

Assuming the above configuration, the following can be determined

LgSvrl

AvaιlableCapacιty=100 CurrentLimits = j ShrMemSeg=10, Semaphores=20, Processors=12 }

LgSvr2

AvaιlableCapacιty=100

CurrentLimits = { ShrMemSeg=10, Semaphores=20, Processoιs=12 }

LgSvr3 AvaιlableCapacιty=100

CuiTentLimits = { ShrMemSeg=10, Semaphores=20, Processoιs=12 }

MedSvrl

A vai lableCapacιty=50

CurrentLimits = { ShrMemSeg=5, Semaphores=10, Processoιs=6 } MedSvr2

AvaιlableCapacιty=50

CurrentLimits = { ShrMemSeg=5, Semaphores=10, Processoιs=6 (

MedSvr3

AvaιlableCapacιty=50 CιπrentLιmιts = { ShrMemSeg=5, Semaphores=10, Processoιs=6 |

MedSvr4

AvaιlableCapacιty=50

Cm rentLimits = { ShrMemSeg=5, Semaphores=10, Processoιs=6 } MedSvr5

AvaιlableCapacιty=50

CurrentLimits = ' ShrMemSeg=5, Semaphores=10, Processors=6 J

Failui e Scenai to

The configuration above shows FaιlOverPohcy=Load and SystemZones The database zone

(System Zone 0) is capable of handling up to two failures Each server has adequate Limits to support up to three database application groups (with an expected performance drop when all database application groups are running on one server) Similarly, the application zone has excess capacity built into each system

In this example, each of MedSvrl through MedSvr5 specifies Limits to support one database, even though the application groups G4 through G8 do not specify Prerequisites This configuration allows a database to fail across SystemZones if absolutely necessary and run on the least loaded application zone machine

For the first failure example, assume system LgSvτ3 fails The cluster management application scans all available systems in Database2's SystemList, with the same SystemZones grouping as LgSvr3 The cluster management application then creates a subset of systems meeting the application group's Prerequisites In this case, LgSvrl and LgSvr2 meet all necessary Limits, and Database 1 is brought online on LgSvrl The following configuration for the database zone is produced

LgSvrl

AvailableCapacity=0

CurrentLimits = { ShrMemSeg=5, Semaphores=10, Pιocessors=6 }

LgSvr2

AvaιlableCapacιty= 100 CurrentLimits = { ShιMemSeg=10, Semaphores=15, Processors=12 }

In this scenario, a further failure of a database can be tolerated, as each system has remaining Limits sufficient to accommodate the database application group miming on the peer system

Cascading Failure Scenai to

If the performance of a specific database is unacceptable with two database groups running on one server (or three following a second failure), the SystemZones policy has another helpful effect Failing a database group into the application zone has the effect of resetting its preferred zone For example, in the above scenai 10, Database 1 has been moved to LgSvrl The administrator could reconfigure the application zone to move two application groups to one system Then the database application can be switched to the empty application server (MedSvrl -MedSvrS) This will place Databasel in Zonel (the application zone) If a failure occurs in Databasel , the least-loaded server in the Application zone meeting its Prerequisites is selected as the failover target

System Suitable for Implementing the Present Invention

Fig 17 depicts a block diagram of a computer system 10 suitable for implementing the present invention Computer system 10 includes a bus 12 which interconnects major subsystems of computer system 10 such as a central processor 14, a system memory 16 (typically RAM, but which may also include ROM, flash RAM, or the like), an input/output controller 18, an external audio device such as a speaker system 20 via an audio output interface 22, an external device such as a display screen 24 via display adapter 26, serial ports 28 and 30, a keyboard 32 (interfaced with a keyboard controller 33), a storage interface 34, a floppy disk drive 36 operative to receive a floppy disk 38, and a CD-ROM drive 40 operative to receive a CD-ROM 42 Also included are a mouse 46 (or other point-and-chck device, coupled to bus 12 via serial port 28), a modem 47 (coupled to bus 12 via serial port 30) and a network interface 48 (coupled directly to bus 12)

Bus 12 allows data communication between central processor 14 and system memory 16, which may include both read only memory (ROM) or flash memory (neither shown), and random access memory (RAM) (not shown), as pieviously noted The RAM is generally the main memory into which the operating system and application programs are loaded and typically affords at least 16 megabytes of memory space The ROM or flash memory may contain, among other code, the Basic Input-Output system (BIOS) which controls basic hardware operation such as the interaction with peripheral components Applications resident with computer system 10 are generally stored on and accessed via a computer readable medium, such as a hard disk drive (e g , fixed disk 44), an optical drive (e g , CD-ROM drive 40), floppy disk unit 36 or other storage medium Additionally, applications may be in the form of electronic signals modulated in accordance with the application and data communication technology when accessed via network modem 47 or interface 48

Storage interface 34, as with the other storage interfaces of computer system 10, may connect to a standard computer readable medium for storage and or retπeval of information, such as a fixed disk drive 44 Fixed disk drive 44 may be a part of computer system 10 or may be separate and accessed through other interface systems Many other devices can be connected such as a mouse 46 connected to bus 12 via serial port 28, a modem 47 connected to bus 12 via serial port 30 and a network interface 48 connected directly to bus 12 Modem 47 may provide a direct connection to a remote server via a telephone link or to the Internet via an internet service provider (ISP) Network interface 48 may piovide a direct connection to a remote server via a direct network link to the Internet via a POP (point of presence) Network interface 48 may provide such connection using wireless techniques, including digital cellular telephone connection, Cellular Digital Packet Data (CDPD) connection, digital satellite data connection or the like Many other devices or subsystems (not shown) may be connected in a similar manner (e g bar code leaders, document scanneis, digital cameras and so on) Conversely, it is not necessary for all of the devices shown in Fig 17 to be present to practice the piesent invention The devices and subsystems may be interconnected in different ways from that shown in Fig 17 The operation of a computer system such as that shown in Fig 17 is readily known in the art and is not discussed in detail in this application Code to implement the present invention may be stored in computer-readable storage media such as one or more of system memory 16, fixed disk 44, CD-ROM 42, or floppy disk 38 Additionally, computer system 10 may be any kind of computing device, and so includes personal data assistants (PDAs), network appliances, X-window terminals or other such computing devices The operating system provided on computer system 10 may be MS-DOS®, MS-WINDOWS®, OS/2®, UNIX®, Linux® or other known operating system Computer system 10 also supports a number of Internet access tools, including, for example, an HTTP-compliant web browser having a lavaScπpt interpreter, such as Netscape Navigator® 3 0, Microsoft Explorer® 3 0 and the like

Moreover, regarding the messages and/or data signals described herein, those skilled in the art will recognize that a signal may be directly transmitted from a first block to a second block, or a signal may be modified (e g , amplified, attenuated, delayed, latched, buffered, inverted, filtered or otherwise modified) between the blocks Although the signals of the above described embodiment are characterized as transmitted from one block to the next, other embodiments of the present invention may include modified signals in place of such directly transmitted signals as long as the informational and or functional aspect of the signal is tiansmitted between blocks To some extent, a signal input at a second block may be conceptualized as a second signal derived from a first signal output from a first block due to physical limitations of the circuitry involved (e g , there will inevitably be some attenuation and delay) Therefore, as used heiein, a second signal derived from a first signal mcludes the first signal or any modifications to the first signal, whether due to circuit limitations or due to passage through other circuit elements which do not change the infoimational and/or final functional aspect of the first signal

Other Embodiments

The present invention is well adapted to attain the advantages mentioned as well as others inherent therein While the present invention has been depicted, described, and is defined by reference to particular embodiments of the invention, such references do not imply a limitation on the invention, and no such limitation is to be mfeπed The invention is capable of considerable modification, alteration, and equivalents in form and function, as will occur to those ordinarily skilled in the pertinent aits The depicted and described embodiments are examples only, and are not exhaustive of the scope of the invention

The foregoing described embodiments include components contained within other components It is to be understood that such architectures are merely examples, and that in fact many other architectures can be implemented which achieve the same functionality In an abstract but still definite sense any arrangement of components to achieve the same functionality is effectively "associated" such that the desired functionality is achieved Hence, any two components herein combined to achieve a particular functionality can be seen as "associated with" each other such that the desired functionality is achieved, irrespective of architectures or intermediate components Likewise, any two components so associated can also be viewed as being "operably connected," or "operably coupled," to each other to achieve the desired functionality

The foregoing detailed description has set forth various embodiments of the present invention via the use of block diagrams, flowcharts, and examples It will be understood by those within the art that each block diagram component, flowchart step, operation and/or component illustrated by the use of examples can be implemented, individually and/or collectively, by a wide range of hardware, software, fn mware, or any combination thereof

The present invention has been described in the context of fully functional computer systems, however, those skilled in the art will appreciate that the present mvention is capable of being distributed as a program product in a variety of forms, and that the present invention applies equally regardless of the particular type of signal bearing media used to actually carry out the distribution Examples of signal bearing media include recordable media such as floppy disks and CD-ROM, transmission type media such as digital and analog communications links, as well as media storage and distribution systems developed in the future

The above-discussed embodiments may be implemented by software modules that perform certain tasks The software modules discussed herein may include script, batch, or other executable files The software modules may be stored on a machine-readable or computer-readable storage medium such as a disk drive Storage devices used for storing software modules in accordance with an embodiment of the invention may be magnetic floppy disks, hard disks, or optical discs such as CD- ROMs oi CD-Rs, for example A storage device used for storing firmware or hardwaie modules in accordance with an embodiment of the invention may also include a semiconductor-based memory, which may be permanently, removably or remotely coupled to a microprocessor/memory system Thus, the modules may be stored within a computer system memory to configure the computer system to perform the functions of the module Other new and various types of computer-readable storage media may be used to store the modules discussed herein

The above descnption is intended to be illustrative of the invention and should not be taken to be limiting Other embodiments within the scope of the present invention aie possible Those skilled in the art will readily implement the steps necessary to provide the stmctures and the methods disclosed herein, and will understand that the process parameters and sequence of steps are given by way of example only and can be varied to achieve the desired structure as well as modifications that are within the scope of the invention Variations and modifications of the embodiments disclosed herein can be made based on the description set forth herein, without departing from the scope of the invention

Consequently, the invention is intended to be limited only by the scope of the appended claims, giving full cognizance to equivalents in all respects.