Automatic discovery of physical connectivity between power outlets and it equipment

0001The present invention generally relates to the field of power management in a data center, and more specifically, the operation of a data center having an automatic detection of connectivity between a power outlet and an IT device and an automatic connectivity discovery function. Regarding the method.

0002Intelligent power distribution devices provide improved power distribution and monitoring capabilities for certain sensitive electrical and electronic applications. An exemplary application that demonstrates the usefulness of deploying intelligent power distributors is to power multiple computer servers involved in providing network services on a predefined schedule based on power management policies. In supply. Here, the ability to control and monitor power distribution is an immeasurable tool for computer network operators and IT personnel, and for use in comprehensive power optimization.

0003One of the above types of intelligent power equipment is the Dominion PX Intelligent Power Distribution Unit (IPDU) developed and sold by Raritan Corp. in Somerset, NJ, USA. Unit). The Dominion PX IPDU provides increased operational and monitoring capabilities at each of the AC power outlets contained in the device. In general, these features include the ability to turn an outlet on and off, among other features, and also provide a power consumption measurement for that outlet. It is desirable for the intelligent power device or the equipment that monitors the intelligent power device to know which specific equipment is plugged into each outlet of the intelligent power device on the other side of the power cable.

0004In addition, network administrators are often required to maintain a data center power connectivity topology. One way to maintain a power connectivity topology is with spreadsheets or in a centralized configuration database. This will be updated by the network administrator from time to time. Other data center asset management systems are also available to track physical power connectivity relationships. It relies on manual entry of physical connections using serial numbers in barcode readers and nameplates. Once entered, the data can be submitted to the topology rendering engine. The topology rendering engine can submit the topology as a report or as a topology map for intuitive visualization. In large data centers, which can contain thousands of servers, manually maintaining the data center power topology is a daunting and error-prone task.

<p num="0005"> Nevertheless, the importance of maintaining an accurate and up-to-date power topology is increasing in the area of network management and operation. As the cost of computing goes down, the cost of power usage by the data center becomes a cost factor. Therefore, reducing power consumption is a concern for network administrators. Similarly, recent Green Initiatives are providing incentives to reduce power usage in data centers. Organizations like the Green Grid publish data center energy efficiency indicators. Data centers measure themselves against these indicators when assessing efficiency. All of these data center management requirements benefit from highly accurate data center power topologies.</p><p num="0006"> Certain auto-discovery topology tools for networks are known. These tools like ping, tracert and mping disclose a logical connectivity map for your network. However, they do not provide an automatic discovery of the physical connectivity between IT equipment and power outlets. Currently, the only way to determine which equipment is attached to a particular outlet of a power distributor is to enter that information manually.</p>

<p num="0007"> Systems and methods based on the principles of the present invention automatically discover physical connectivity topologies for information technology (IT) equipment in data centers. Topology shows the connection between IT equipment and a power outlet. Systems based on the principles of the present application apply a set of heuristics to identify candidate power outlets for individual servers or other IT equipment. On one side, for a particular facility, candidate outlets are selected based on their physical proximity to the IT facility. These candidates are sequentially and iteratively narrowed down based on theoretical power consumption data, actual power consumption data, CPU utilization, and correlation of state change events.</p><p num="0008"> The physical position can be determined using a variety of techniques such as ultrasound detection or RFID. This information can then be used to increase the physical connectivity between the server and the power outlet. In a typical situation, power consumption data provided by an IT equipment vendor can be used to narrow down candidate outlets by systematically comparing outlets within the operating range provided by that vendor. This nameplate data typically exceeds the actual power consumption and may not narrow the candidate outlets to a definitive mapping. In these cases, the actual data can further narrow down the candidate outlets. CPU utilization data about the server can be collected over a period of time and quantized to mitigate noise and other artifacts. The actual power consumption over the same time period is collected from the candidate power outlets using the appropriate IPDUs. Pattern matching between quantized CPU utilization and power consumption graphs identifies a match. In addition, state changes reflected in power and CPU utilization data further narrow down candidate power outlets for a given IT facility. Quantized CPU utilization and power consumption data can also be used for these comparisons. If heuristics narrow down the candidates but do not converge, the administrator can look at utilization graphs and other data outputs and make subjective conclusions about the best outlet candidates for an IT facility.</p><p num="0009"> A system and method for providing an automatic identity association between an intelligent power distribution unit outlet and a target device such as a computer server powered by that outlet is a power outlet. It can include a power management unit or a power distribution unit that implements data collection in. The IT equipment's power requirement profile specified by the equipment vendor as well as the actual usage patterns measured over time correlate with the power consumption patterns detected at the candidate power outlets. Further correlation is made between the time sequence of turning the server on and off, changing the computing workload of the server, and certain state changes on IT equipment such as virtual machine migration. These state changes can be detected by the monitoring system and are reflected in the actual changes in power utilization at the power outlets. Heuristic rules and indicators are applied sequentially and iteratively until the number of candidate power outlets matches the number of power supply units for the IT equipment.</p><p num="0010"> The discovery of physical connectivity topologies based on the principles of the present invention maintains a high degree of integrity. In addition to key indicators such as actual CPU utilization and power consumption, other indicator characteristics of a particular functionality of a given IT facility can further identify candidate power outlets. In addition, the interface can be used to allow administrators to verify power matching by actual visual inspection of CPU utilization and power consumption graphs for IT equipment and discovered power outlets. ..</p>

0011<figref num="1">It is a figure which shows the system based on the principle of this invention.</figref><figref num="2">It is a figure which shows another system based on the principle of this invention.</figref><figref num="3">It is an exemplary graph for implementing aspects of a heuristic rule based on the principles of the present invention.</figref><figref num="4">Other exemplary graphs for implementing aspects of heuristic rules based on the principles of the present invention.</figref><figref num="5">It is an exemplary graph for a single intelligent power unit over a 24-hour period, based on the principles of the present invention.</figref><figref num="6">It is an exemplary graph of CPU and power utilization for a single intelligent power unit over a period of 3 hours, based on the principles of the present invention.</figref><figref num="7">FIG. 6 illustrates an exemplary graph of a CPU and a processed view of data for a single intelligent power unit over a period of 3 hours, based on the principles of the present invention.</figref><figref num="8">It is a figure which shows the exemplary histogram transformation of the PDU utilization at the socket level based on the principle of this invention.</figref><figref num="9">It is an exemplary flow diagram of the automatic association framework based on the principle of the present invention.</figref><figref num="10">It is an exemplary flow diagram of an automatic association algorithm based on the principle of the present invention.</figref>

0012FIG. 1 discloses a system 100 based on the principles of the present invention. System 100 includes N racks of IT equipment, of which three racks 102, 104, and 106 are shown. These are the types that are typically used in data centers. These racks can hold any number of different types of IT equipment, including servers, routers and gateways. As an example, rack 102 shows two vertically mounted power strips 114, 116, each of which contains eight power receptacles, on those power strips. The power supply for IT equipment is physically connected. Other racks in the data center have similar power outlet units, and those putter outlet units can be mounted in a variety of configurations.

0013In this exemplary System 100, these power strips are of a type capable of providing power consumption data and other functionality, such as the Dominion PX provided by Raritan, Somerset, NJ, USA. Alternatively, these units can be referred to as power distribution units or PDUs. These power distribution units provide TCP / IP access to power consumption data and outlet level switching, via SNMP and email for events such as threshold crossing or single on / off power cycling. Can provide alerts. The PDU integrates with a wide variety of KVM switch solutions such as the Dominion KX2 and Paragon II KVM switches offered by Raritan. Racks 104 and 106 may be similarly equipped. PDUs are often highly configurable, and these exemplary power distribution units 114, 116 are power managers (Power). Has a direct interface with Manager) 108. The power manager 108 may be an element management system capable of configuring multiple IPDUs in a power distribution network. Power managers can also collect IT utilization information provided by IPDUs. An exemplary power manager 108 may be equipped to provide remote access to the administrator 112 and can address power distribution units 114, 116 through Internet Protocols. The power manager 108 can be configured to discover and aggregate the data in the Database 110 that provides the data for heuristics applied under the principles of the present invention. As will be explained later, this data includes actual power consumption data, IT equipment specifications, CPU utilization data, theoretical power consumption data, and state change events on IT equipment.

0014Figure 2 shows another exemplary system 200 with a data center containing N racks of IT equipment. Three racks 220, 222, and 224 accessible to administrator 208 through IP network 212 are disclosed for illustrative purposes. Rack 224 includes IT equipment as well as a power distribution unit with intelligent power functions. Among these features is the collection of data such as actual power consumption data at the output outlet level. Racks 220 and 222 are similarly equipped and include an environmental sensor 228 that can operate to detect environmental conditions in the data center. The optional power data aggregator 226 has a power distribution unit and an interface to aggregate the data from the outlets. Some of these racks 220, 222, 224 are also equipped with sensors and circuits to determine their physical proximity to the power outlets. Rack-mounted sensors can be monitored by IPDUs to estimate the amount of power dissipation due to increased temperature. The amount of temperature rise directly correlates with the amount of power consumption exercised by the server and can therefore be used in said correlation. System 200 includes an authentication server 214 and a remote access switch 204, such as the Dominion KX KVM over IP switch, which interfaces with the administrator 208.

0015The switch 204 is further interconnected with a data store 202 for storing and retrieving data useful in determining the physical connectivity of the IP equipment to the power outlets. This data includes, but is not limited to, power outage reference signatures, theoretical power signatures, actual power signatures, actual power data and other associations. In addition, the power distribution manager 206 interfaces with the KVM switch 204. The KVM switch 204 allows the administrator 208 to remotely access power distribution unit data from various power distribution units located on racks 220, 222, 224. Another database 216 is accessible through IP network 212 to store physical location data related to IT equipment and power outlets. The change alert server 218 is also optionally connected and accessible through KVM switch 204. In operation, data from racks and power distribution units in the data center is collected and stored through the IP network and is selectively accessible to administrator 208. The power distribution center and reporting facility access the data and perform the methods according to the invention to identify the physical connectivity between the IT facility and the power outlet. KVM switch 204 can be used to actively connect to the server to be associated. This is because it increases the usage on the server. Administrators can use this KVM approach to improve connectivity discovery on selected servers that can provide similar power signatures in normal operation.

0016In each of the above systems 100 and 200, power distribution units and KVM switches and / or other administrator equipment or servers store data in a database for later use and to apply correlation heuristics. Programmed to collect. The data obtained through monitoring can be divided into two main categories. One is time series information that gives the value of data at an arbitrary time point. The second is a time-stamped event that affects both IT and power systems. Examples of the latter include rebooting the server machine and starting up the server. Among the various data attributes useful in the principle-based correlation method of the present invention are the data related to the theoretical power usage requirements of a particular IT facility, the practice at a particular power outlet, measured over time. There is power consumption data, actual CPU utilization data about the servers in the data center collected over time, and the physical distance relationship between the identified server and the identified power outlet. In addition to this data, other useful characterization data can be obtained and stored in the data store. This data can include data characteristics for a particular type of IT equipment found in a data center. For example, email servers, web servers, routers, etc. often have identifiable characteristics depending on their specific use in the data center, which is temperature, CPU utilization, on-to-off state. Changes in, including any other characteristics that can be identified alone or in combination with other server characteristics.

0017The correlation engine can be implemented on either a power management unit, a general purpose computer, or a dedicated server that has access to the data store to run any heuristics and to deploy connectivity maps for the entire data center. it can. When heuristics are applied, the number of outlets that can connect to a particular possible server is narrowed down and typically converges to the identified outlet for that server. If heuristics are applied but the possible candidates cannot be reduced to a single response, the administrator can access and identify a graphic representation of a particular characteristic, such as a CPU utilization graph, power consumption graph, etc. You may subjectively assess the likelihood that your server is physically connected to a particular outlet. Databases and rendering engines can be implemented using known data structures and rendering software so that the topography of the physical connectivity of the data center can be rendered.

0018Any particular heuristic is optional and additional heuristic rules and indicators can be added to the process for identifying the physical connectivity between the server and the outlet. In one exemplary method, a set of power outlets is identified as a probable candidate for specific IT advice. These probable candidates can be based on previously given connectivity data, association clustering, physical location or best possible candidates entered by the data administrator. The additional information aids convergence by matching a likely pair of unknowns, rather than applying the decision to a pair of completely unknown powers and IT endpoints. For these candidates, a set of heuristic rules is applied in an attempt to map IT equipment to one or more specific outlets. The heuristic process is completed when the number of candidate power outlets matches the number of power supply units on the IT equipment, or when all heuristics are exhausted. If all heuristics are exhausted, the administrator may make subjective choices based on looking at the data of the remaining candidate outlets.

0019Some indicators that can be used in the heuristic process are name plates. values), the actual power consumption pattern, the time sequence of IT equipment state change events, and the physical location of the IT equipment with respect to the power outlet. So, for example, assuming a set of 20 candidate power outlets for a given IT facility, a subset is erased because it is not within a physical distance of the IT facility. This indicator reinforces the typical practice of locating a server within a specified maximum distance from its outlet. Nameplate information is used to group servers by their average power consumption level, and pattern matching algorithms match selected subsets of servers and electrical power only if the power values overlap. Outlets can be determined. For example, if the power outlet is delivering M watts of power and the server has a maximum nameplate power of N watts, then if M &gt;&gt; N, then the correlation between the power outlet in question and the server is Absent. Of the remaining candidate outlets, heuristics are applied to identify the actual CPU utilization and correlate it with the actual power consumption at that power outlet. This reduces the number of candidate outlets to the identified pairs. If not reduced, additional heuristics are applied to determine the actual state change reflected in the CPU utilization and power consumption graphs. Additional heuristics can be applied by analyzing IT utilization throughout the day using histograms. Time series data can be transformed into other regions in the frequency or spatial region to improve correlation within the context of power characteristics.

0020In one aspect of the invention, a first candidate for a potential outlet for a particular server is identified through IP addressing. The number of IPUs in electricity distribution can be found using various methods based on their function. In the case of Raritan DPX, IPMI discovery provides sufficient information about the existence and configuration of these units. Similarly, network management technology provides the ability to discover server system details, including network IP addresses, that can be used to monitor and measure IT utilization over the network. Data is collected from the server and from the power outlet unit using the IP address. The data is aggregated in the data store. Data collection methodologies available for the proposed invention include SNMP, IPMI, WMI and WS-MAN. All these standard management interfaces provide useful remote monitoring capabilities for the present invention. The data can be time stamped. It allows power usage, CPU usage and events to be correlated between different candidate power outlets and different IT equipment.

0021A, B, and C of FIG. 3 show three exemplary graphs 302, 304, 306 exemplifying one aspect of heuristics that can be applied on the basis of the principles of the present invention. Graph 302 in A of FIG. 3 shows CPU utilization (Y-axis) over time (X-axis). CPU utilization data is raw, non-quantized data that represents all the cores of the candidate IT equipment you are considering. This unquantized data is somewhat noisy and may not be optimal for correlation with other data. Graph 304 in B of Figure 3 shows the same data quantized to eliminate artifacts and noise. In this example, the values used are approximately quantized to the integer values 1 and 2. However, other quantization methods can be used without departing from the principle of the present invention. Again, the usage data corresponds to all cores for the candidate IT equipment. Graph 306 in C in Figure 3 shows the actual power consumption of the candidate outlets over the same time period. Time is tracked using the time stamps added during the data collection. There are event changes that illustrate changes in CPU utilization, as indicated by arrows 308 and 310. Similarly, in power graph 306, the data reveals a power spike at 312. This spike 312 may correlate with events 308, 310 in core utilization for unquantized and quantized graphs. Event timestamp comparison is another data indicator that can be used to correlate this candidate IT facility with candidate power outlets.

0022FIG. 4 shows exemplary utilization data graphs 402, 404 and corresponding histograms 406, 408 that can be used to correlate candidate power outlets with IT equipment in heuristics based on the principles of the invention. Graph 402 represents raw utilization data for all cores of an IT facility throughout the day. Here, the utilization value falls between about 0 and about 100. This raw utilization data is not easy to mine to get indicators that can be used to correlate with candidate power outlets. Utilization histogram 406 classifies utilization based on the frequency of utilization of a particular selected value. Therefore, the histogram depicts how often IT equipment was used at a particular level during a given period of time.

0023Graph 406 details how often the processor cores of a given IT equipment switch to different utilization levels. In this example, graph 404 is obtained by halving the raw utilization data over a given period of time. The graph is normalized around 0 on the vertical axis because graph 404 shows the change from the current utilization state to lower or higher utilization. Histogram 408 is an analysis showing the frequency of changes in utilization on the X-axis versus the frequency of use on the Y-axis. This data provides similar graphic histograms and spectra for candidate power outlets, and then the correlation techniques of the invention are examined by examining them using computer-implemented power matching or manually if necessary. Can be used in.

0024Figure 5 shows a power utilization graph for a single Dominion PX over a 24-hour period. Data 502 shows the power utilization of one of the sockets, which is reduced to 0 at a particular time 501 corresponding to CPU utilization 0 (or unavailable). If events such as power recycling and shutdown do not exist at the same time (not at the same time in Figure 5), it is unlikely that an event-based correlation will be achieved. Available PDUs are not currently equipped with event logging capabilities for individual sockets in the PDU. PDUs based on the principles of the invention extend such logging for the purpose of correlating events between servers and PDU sockets. Since the order of power recycling dominates, the delay required to make the association between the server and the PDU is achievable.

0025Figure 6 shows an example of CPU and power utilization over a 3-hour period. Data 601 represents CPU utilization over a 3-hour period. In this exemplary embodiment, the sum of all processor cores in a particular server includes all four cores in this processor. Therefore, the total value demand is divided by 4 to express the utilization as a ratio of electricity. Data 602 represents the power utilization for the server over the same period logged by a PDU. CPU utilization and power increase steadily over time, as evidenced by data 601 and 602. As can be seen by data 602, the server consumes an average of 178 watts for the average CPU utilization of 27.90 indicated by data 601.

0026Figure 7 shows an example of CPU utilization and a corresponding histogram of the processed data, highlighting the low utilization of the server. Data 701 shows server activity and how active the server is over a given time period. According to the principles of the present invention, as can be seen from the data 701, the conversion of time series information from server utilization or PDUs can be useful when correlating based on data values. Data 702 is an example of a histogram-based approach to transforming time series data 701 into utilization context. Histogram data 702 may be correlated with a histogram of PDU utilization based on the principles of the invention.

0027Figure 8 shows an exemplary PDU utilization and corresponding histogram view for a single outlet. Data 801 represents the power in watts of a given power outlet over a given time period. As shown, the average power at the outlet is 137.27 watts. Data 802, represented by a histogram transformation of PDU utilization at the socket level, shows that most of the power activity at the socket level corresponds to average power consumption over the same given time period.

0028FIG. 9 shows an exemplary flow diagram 900 of a heuristic auto-association framework based on an embodiment of the present invention. Once started, step 901 gets the environmental components of the system. Specifically, in step 901, the auto-association framework collects configuration configuration information about servers and PDUs in the system, and in step 902 downloads the configuration configuration information for storage. Step 903 determines if all configuration configuration information has been collected. If there is additional configuration information to collect, steps 901 and 902 are repeated until the process is complete. Utilization measurements from identified servers and PDUs are collected during step 904 and stored in the database at step 906. Steps 904 and 906 are repeated until terminated by the user in step 905.

0029FIG. 10 shows an exemplary flow diagram 1000 of a heuristic auto-association algorithm based on the principles of the present invention. At step 1001, the system determines if the server asset information is available for analysis. If the information is available, in step 1002 the data is filtered based on the server maximum and average power information. The filtered information from step 1002 and the utilization data stored in step 906 of FIG. 9 are passed for analysis in step 1003. During step 1003, derived metrics (ie sum, histogram, maximum and minimum) from utilization data from servers and PDUs are calculated. Similarly, in step 1004, event analysis is performed to detect the timing of specific events on various PDUs and servers and group them based on their relative occurrence. It may be based on server asset information from various server manufacturers supplied by database 1011 and entered in step 1004. The analyzed data from steps 1003 and 1004 are passed through first-level heuristics in step 1005. During step 1005, servers and PDUs are grouped into pairs based on the data and / or event matching. During step 1006, it is determined whether the pairing from step 1005 is the correct association between the server and the PDU. If determined to be correct, in step 1007 that information is passed to and stored in the server and PDU association databases. If it is not determined that the server PDU association in step 1005 was determined by step 1006, the process proceeds to step 1008 and uses second-level metrics (ie, detailed wavelets, processor characteristics, quantization, etc.) to server PDUs. Further classify the pair. Su Tep 1009 performs higher levels of heuristics and attempts to group servers and PDU devices based on a second metric and classification. If the association is determined to be correct in step 1006, the server PDU association information is stored in the database in step 1007. The algorithm terminates once it is determined through step 1012 that all servers have been associated with all PDUs.

0030These and other aspects of the invention can be implemented in existing power management topologies. Data acquisition capabilities for integrating CPU utilization, actual power utilization, nameplate specifications and other data are currently known and in use. Data related to assets can be retrieved from the vendor list or imported from corporate asset management tools. Basic data schemes may be used to aggregate data, including tables or hierarchical data structures. Heuristic processes can be implemented on a general purpose computer or can be a separate feature implemented within an existing power management unit. Rendering engines with front-end interface capabilities for rendering graphs and / or interfaces are also known in the art.