System and method for resource allocation in a multi-domain network

Field of the Invention

0001The present invention relates to a technical field of network communication, particularly to a system and a method for implementing resource allocation in network communication.

Background of the Invention

0002As network communication technology develops, usually there are special transmission demands for specific services in network communication. Therefore, it is necessary to control network resources to meet the transmission demands of specific services, e.g., restrict File Transfer Protocol (FTP) bandwidth on the backbone network, provide higher priority for database access, realize Internet Service Providers (ISP) providing different transmission for voice, video, and other real-time services and provide bandwidth and low time delay assurance for time-sensitive multimedia services.

0003To this end, Internet Protocol Quality of Service (IP QOS) technology appeared. IP QOS refers to the capability of providing services over IP networks, i.e., provide required service capability for specific flows across under-layer IP networks including Frame Relay (FR), Asynchronous Transfer Mode (ATM), Ethernet, and Synchronous Digital Hierarchy (SDH); usually, the technical criteria used to measure IP QOS include:

0004bandwidth/throughput: which refers to the mean velocity of flows for specific application between two nodes;

0005time delay: which refers to the average round-trip time of a data packet between two nodes in the network;

0006dithering: which refers to variation of the time delay;

0007loss rate: which refers to the percentage of lost packets during transmission in the network and is used to measure the capability of the network in forwarding subscriber data correctly;

0008availability: which refers to the percentage of the available service time during which the network can provide service for subscribers.

0009To implement the end-to-end QOS function in the communication network, usually each network element, e.g., router, Ethernet switch, etc., shall have the capability of classifying messages and providing different processing, queue management, and dispatching for different classes of messages to meet different QOS requirements of different services, as well as providing traffic supervision and control, traffic shaping, adjusting message output speed, and determining whether to allow the subscriber's data flow to use network resources.

0010IP QOS includes Resource Reserve Protocol-based (RSVP-based) Integrated Service (IntServ), which is an end-to-end flow-based QOS technology; with that technology, before communication, the applications at two ends shall set up an end-to-end communication path with out-of-band RSVP signaling according to class of service and the requirement for network resources, and each router along the path shall record state information of each flow and provide corresponding service assurance.

0011To seek for expansibility and simplicity, IETF organization has put forward Difference Service (DiffServ) technology, which is a class-based QOS technology and is mainly used in the backbone network; the technology classifies services, performs flow control, and sets DSCP (Difference Service Code Point) domain of the message at the network ingress according to requirements of services; it distinguishes communication types in the network according to the predefined QOS mechanism and provides policies, including resource allocation, queue dispatching, and packet discarding, for Per Hop Behave (PHB) ; all nodes in the DiffServ domain will obey PHB according to the DSCP field in packets.

0012To incorporate the advantages of above IntServ and DiffServ, Integrated Service over Difference Service (IntServ over DiffServ) model is developed, which presumes the two ends of the communication network support IntServ/RSVP and regards some domains that don't support IntServ/RSVP along the end-to-end path, e.g., DiffServ domain. The DiffServ domain is regarded as a virtual connection in the IntServ domain; flow-based RSVPs are transmitted transparently in the DiffServ domain to the other end; certain bandwidth is reserved between the two ends. As shown in <figref idref="f0001">Fig.1</figref>, the bandwidth from the source end to ingress of the DiffServ domain and from egress of the DiffServ domain to the destination end is guaranteed; however, in the DiffServ domain, the bandwidth is only guaranteed for aggregate flows but may not be guaranteed for individual flows. To overcome the phenomenon, the DiffServ domain shall support Aggregate Resource Reserve Protocol (RSVP).

0013The aggregate RSVP collects flow-based RSVP requests at ingress node of the DiffServ domain, aggregates the requests, and then requests for a total bandwidth from the egress node of the DiffServ domain. When there is a new RSVP request or a Cancel request, the aggregate RSVP adjusts the reserved bandwidth between edges of the DiffServ domain. Since the total bandwidth is exactly the sum of the bandwidths of individual flows, end-to-end QOS assurance can be provided.

0014Even though IntServ over DiffServ technology can implement reasonable allocation of communication network resources, it has the following disadvantages: <ol id="ol0001" compact="compact"><li>(1) the technology is only applicable to IP network and is tightly coupled with RSVP; it applicability is limited;</li><li>(2) the technology is only suitable for aggregate model but is not suitable for peer-to-peer model;</li><li>(3) the resulting aggregate flow from data flows is solely determined by the network, and there is no interaction with the application terminals; therefore, the network has to maintain and manage a large quantity of resource allocation policy information;</li><li>(4) it has no aggregate bandwidth pre-allocation mechanism; instead, it employs first application flow-triggered establishment mechanism;</li><li>(5) there is no aggregate bandwidth allocation mechanism between the host and the Aggregator/De-Aggregator; in Client/Server model, the Aggregator/De-Aggregator connected with the server has to support a large amount of data flow-based queues;</li><li>(6) it has no resource preemption mechanism.</li></ol>

0015As Multi-protocol Label Switch (MPLS) technology emerges, people begin to try to solve QOS problem of traffic transmission with MPLS, including MPLS DiffServ and MPLS TE. The combination of MPLS and DiffServ is referred to as MPLS DiffServ (or MPLS CoS), which means to aggregate DSCPs or labels at edge of the network and process DSCP-based PHB or label-based forwarding at core of the network. While MPLS TE utilizes LSP supporting route presentation capability and pilot network traffic on the premise of limited network resources so as to match the actual network traffic load with the physical network resources and thereby improving QOS of the network.

0016DS-Aware TE is a MPLS-based indirect QOS technology, which optimizes network resource utilization through reasonable resource configuration and effective control of routing process. Since MPLS incorporates Lay-2 and Lay-3 technologies, it has intrinsic advantages in solving traffic engineering. MPLS traffic engineering optimizes network performance through searching for possible paths that meet the requirements of traffic relay with a set of route-constraining parameters. During setup, MPLS LSP carries some constraint conditions, e.g., bandwidth, affiliation property, etc. and sets up a satisfactory path through calculating restricted routes.

0017However, as shown in <figref idref="f0001">Fig. 2</figref>, if the sum of all Expedite Forwarding-type (EF)-type LSP bandwidths exceeds 50% of the total bandwidth, the time delay will exceed M1 ms, which means the traffic transmission requirements can't be met. Therefore, the EF service bandwidth shall be controlled strictly within 50%. As shown in <figref idref="f0002">Fig. 3</figref>, suppose the bandwidth between A and E is 200M, the bandwidths between E and C, between C and D, between D and F, as well as between F and E are all 100M, and a 20M EF flow has been established on the path through A, E, F and D through calculation of restricted paths; if another 20M EF-type flow from A to D is to be established, but it is found through calculation at point A that the bandwidth on the path through A, E, F and D meets the requirement and is optimal; because it is unable to know whether the sum of LSP bandwidth for EF flow established at point E has exceeded 50% of the total bandwidth during the calculation of restricted routes, point A only knows there is an 80M bandwidth from point E to point F. Therefore, it is possible that the time delay for EF-type LSP established at point E is not guaranteed. A solution is to expand existing Multi-protocol Label Switch Difference Service-Aware Traffic Engineering (MPLS DS-Aware TE), to inform resource occupation information at point E of point A.

0018The basic idea of MPLS DS-Aware TE is to enhance the class-based constraint condition on the basis of original MPLS TE, i.e., the nodes in the MPLS DS-Aware TE domain disseminate resource and occupation information according to each subscriber-defined service class to each other through Interior Gateway Protocol (IGP); during the connection is set up, MPLS signaling protocol specifies not only bandwidth but also service type, EF or Assured Forwarding (AF), which specifies the type to which the resource pertains to, so that the network resources are utilized to the best according to the subscriber's demands.

0019However, the existing DS-Aware TE still has the following disadvantages: it is only applicable to MPLS network; it only solves the QOS problem in a single autonomous system (AS) but can't solve the QOS problem across ASs; in addition, it can not solve the end-to-end QOS problem that doesn't support multicast resource requests, and thus its applicability range is limited.

0020The document D1 (<nplcit id="ncit0001" npl-type="s"><text>STEFANO SALSANO ET AL: "Inter-domain QoS Signaling: the BGRP Plus Architecture" IETF STANDARD-WORKING-DRAFT, INTERNET ENGINEERING TASK FORCE, IETF, CH, May 2002</text></nplcit>, XP15005141) discloses the BGRP plus architecture for inter-domain resource reservation.

Summary of the Invention

0021An object of the present invention is to provide a system and method for implementing resource allocation in network communication, in order to solve end-to-end QOS problem in network communication.

0022An embodiment of the present invention provides a system for implementing end-to-end resource allocation for data flows in network communication, including:

0023Quality of Service, QOS, domains with separate resource allocation solutions, including a Quality of Service domain connected to a source terminal, a QOS domain connected to a destination terminal and at least one other QOS domain; the source terminal; and the destination terminal; wherein each QOS domain comprises QOS Edge Routers, QERs, and QOS Core Routers, QCRs; wherein data flows flow into a QOS domain through an ingress QER, and data flows output from a QOS domain through an egress QER; wherein the QERs connected with the source terminal, the destination terminal, or the QER of another QOS domain, are adapted for establishing a resource reservation path between the edge of the QOS domain and the source terminal, between the edge of the QOS domain and the destination terminal, or between the edges of the QOS domains, for an aggregate flow generated from data flows from the source terminal, and only the QERs connected with the source terminal and the destination terminal are adapted for maintaining the state of the aggregate flow, while the QERs other than the QERs connected with the source terminal or the destination terminal are adapted for maintaining only edge-to-edge resource reservation for the aggregate flows; wherein the QCRs connected with the QERs in the same QOS domain, are adapted for establishing an intra-domain resource reservation path between the QERs for the aggregate flow; wherein the source terminal is adapted for generating the data flows and sending a resource request message for the data flows to the QERs; and wherein the destination terminal is adapted for receiving the data flows.

0024Another embodiment of the present invention provides a method for implementing class-based end-to-end resource allocation for data flows in network communication based on the system of claim 1, comprising: <ul id="ul0001" list-style="none" compact="compact"><li>establishing a resource reservation path between the edges of the QOS domains for an aggregate flow generated from data flows from the source terminal by the QER connected with the source terminal, the QER connected with the destination terminal, and the QER of another QOS domain;</li><li>maintaining the state of the aggregate flow by only the QERs connected with the source terminal and the destination terminal;</li><li>maintaining only edge-to-edge resource reservation for the aggregate flows by the QERs other than the QERs connected with the source terminal and the destination terminal; and</li><li>establishing an intra-domain resource reservation path between the QERs for the aggregate flow by the QCRs connected with the QERs in the same QOS domain.</li></ul>

0025It can be seen from the above technical solution that the method for managing the communication network through dividing it into a plurality of QOS domains according to the present invention solves the end-to-end QOS problem effectively and has the following advantages: <ul id="ul0002" list-style="none" compact="compact"><li>according to the present invention, end-to-end path information is maintained only at the edge router of the QOS domain connected with source/destination terminal, while the edge routers of other QOS domains maintain only edge-to-edge resource reservation information for aggregate flows, which reduces significantly flow state information maintained at other network nodes as well as overhead of signaling protocol processing and storage at network nodes; in order to reserve resources for aggregate flows, special queues are usually allocated for the aggregate flows; therefore, the required number of queues is reduced significantly when compared with IntServ approach, and thereby the aggregate flow state information at the network nodes is reduced;</li><li>the QOS domains are in a peer-to-peer mode to each other; therefore, the present invention is applicable to both peer-to-peer network model and hierarchical network model;</li><li>as long as IP transmission is used between the two ends, different QOS domains may use different intra-domain resource allocation modes; therefore, the method according to the present invention can facilitate interconnection between different networks, and thereby the applicability of the present invention is expanded;</li><li>the edge nodes obtain data flow identification mode through dynamic protocol; such identification mode only require IP information instead of information on application layer; therefore, the flow classification is simple to implement and needs small configuration workload;</li><li>according to the present invention, aggregate flows with high priority can preempt resources pre-allocated to aggregate flows with low priority, which ensures effective resource utilization and doesn't affect data flows that have been allocated with resources;</li><li>the resource pre-allocation or incremental resource allocation approach reduces effectively the frequency of resource allocation and release as well as the overhead on signaling protocol, and thereby makes the network stable;</li><li>since the resources are allocated on the edge routers at next hops, multicast resource requests is supported and setup of resource reservation path can be further accelerated;</li><li>in most cases, according to the present invention, there are enough resources among edge routers of QOS domains, and thereby it is unnecessary to initiate the intra-domain resource allocation mechanism; instead, only QER processing is required for end-to-end resource allocation; therefore, the mean path setup time is relatively short.</li></ul>

Brief Description of the Drawings

0026<ul id="ul0003" list-style="none" compact="compact"><li><figref idref="f0001">Fig.1</figref> is a schematic diagram of DiffServ-based IntServ model.</li><li><figref idref="f0001">Fig.2</figref> is a resource allocation curve diagram of DS-Aware TE technical solution;</li><li><figref idref="f0002">Fig.3</figref> is a networking diagram of DS-Aware TE;</li><li><figref idref="f0002">Fig.4</figref> is structural diagram 1 of the system according to the present invention;</li><li><figref idref="f0003">Fig.5</figref> is a diagram of procedures of the method according to the present invention;</li><li><figref idref="f0004">Fig. 6</figref> is a diagram of working procedures of the policy server shown in <figref idref="f0002">Fig.4</figref>;</li><li><figref idref="f0004">Fig.7</figref> is structural diagram 2 of the system according to the present invention.</li></ul>

Detailed Description of the Embodiments

0027The present invention solves end-to-end QOS problem for data flows in network communication and can be referred to as Class-Based Integrated Service (CB-IntServ). It combines the control plane of IntServ with the forwarding plane of DiffServ to provide end-to-end QOS assurance in any network, whatever the scale or topology of the network is. In addition, the present invention can be implemented both in individual domains, across domains and across ASs.

0028Providing end-to-end QOS assurance for individual data flows doesn't means steam-based end-to-end paths must be used; instead, for a single network device, what is only required is to reserve enough resources for data flows passing through it, i.e., resource reservation can be aggregated, this is because, in the case of resource reservation, employing multiple QOS queues for the same class of data flows delivers the same effect as employing only one queue substantially; however, in view that different classes of services have different QOS requirements and different traffic models, the resources for them have to be treated differently.

0029In order to determine the quantity of resources to be reserved at a network device, the resource request for data flows has to be sent to every node (i.e., network device) on the transmission path; that function can be implemented through signaling, policy server, or other approaches.

0030Similar to the case of a single network device, for a segment in the network, it is conceivable that the QOS of each data flow passing through that segment can be assured as long as the resources reserved for the class of service among the edges of the segment are enough; likewise, in order to determine the quantity of resources to be reserved in a segment, the resource request for data flows has to be informed of every segment.

0031Usually, the resource reservation comprises two stages, i.e., resource request stage and resource allocation stage; in RSVP, the resource allocation follows the resource request; for a single data service, employing resource pre-allocation will severely waste network resources and result in frequent signaling interaction and affect network stability; however, for an aggregated class (i.e., an aggregate flow), the probability of concurrency of all data flows is very low due to statistical multiplexing; in addition, the number of aggregate flows are limited, it is therefore possible to pre-allocate resources for aggregate flows and increase the quantity of resources dynamically in case that the pre-allocated resources are not enough.

0032In detail, the system for implementing resource allocation in network communication according to the present invention comprises a plurality of QOS Autonomous Systems hereinafter referred to as QOS-AS or QOS domain, each of which is managed separately and a unified QOS policy is used in one QOS domain; said QOS-AS may equal to a route AS or an AREA or not, and the simplest QOS-AS is a device with its interfaces.

0033Internally, a QOS-AS may be implemented through a plurality of protocols, e.g., IP, ATM, FR, MPLS and Ethernet, or with sufficient bandwidth and transparent to the external. Hereunder the present invention is further described, with an IP network as the example. The QOS domain according to the present invention comprises: <ul id="ul0004" list-style="none" compact="compact"><li>QOS Edge Routers (QER), which are located at the edge of a QOS-AS and are connected to application terminals or other QOS-ASs, responsible for classification, labeling, policy, and shaping, etc., for the entire QOS-AS, serving as signaling start point and end point in the QOS-AS, and responsible for establishing resource reservation paths for aggregate flows between the edges and maintaining states of the aggregate flows;</li><li>If a QER is connected with another QOS-AS or an application terminal, resources reservation between the QER and the application terminal or between the QER and a QER of another domain can be performed through a certain signaling protocol; if connected with the application terminal, the QER is called as QOS Application Edge Router (QAER), which stores the states of data flows;</li><li>A QER may belong to two QOS-ASs; in that case, the link on the QER also belongs to the two QOS-ASs, and the QER is virtualized as two QERs;</li><li>If connected with a QOS-AS in a heterogeneous network (e. g. , ATM), the QER is also responsible for mapping QOS parameters between the different networks.</li></ul>

0034QOS Core Router (QCR), which is located in the QOS-AS and is responsible for only forwarding messages, running the signaling protocol in the QOS domain, establishing resource reservation paths for aggregate flows between QERs, and storing state information of the aggregate flows, since the QER has performed classification, labeling, policy, and shaping, etc.;

0035Both QCR and QER are defined with respect to a certain data flow; it is possible that a physical router may be a QCR for some data flows and a QER for some other data flows at the same time.

0036Application terminals, which may be intelligent terminals server, PC, PDA, or serving as s sources or destinations for data flows; an application terminal sends resource request messages to the network via QAER; if the resources meet the requirements, it will perform communication with assured QOS; if the communication is bidirectional, the application terminal is both the sender and the receiver and will request for or reserve resources in both directions; in addition, the application terminal can make the QAER identify it exclusively by a unique identifier, e.g., address or port information sent from the application terminal.

0037Policy server: when an application terminal requests for resources actively to the network, the network will decide whether to allocate resources for it according to a certain policy, e.g., charge, security, etc.; if the number of data flows is small, that function may be implemented on the QAER; however, when scale of the network and subscriber is expanded, policy servers are required to do the work; in addition, the policy server may be active to configure the QERs of QOS domains and to allocate resources for aggregate flows, wherein the policy server allocates resources according to topology information and link resource occupation condition in the QOS domains.

0038The ultimate approach to solve QOS problem in network communication is to assure end-to-end resources (e.g., bandwidth) for data flows; the method for resource allocation in network communication according to the present invention employs technologies including assurance by domain, pre-allocation, and dynamic adjustment (including incremental allocation and statistical adjustment) to solve QOS problem; the method according to the present invention comprises resource allocation solutions in a QOS domain, among QOS domains, and between the application terminal and the network, wherein: <ul id="ul0005" list-style="none" compact="compact"><li>resource allocation solution in a QOS domain: the data flows are aggregated into a limited number of aggregate flows, e.g. , Expedited Forwarding (EF) and Assured Forwarding (AF) defined in DiffServ system, at the QERs of the QOS domain, and the resources are allocated for the aggregate flows from edge to edge in the QOS domain, which is equivalent to establishing a dedicated aggregate flow tunnel between QERs of the QOS domain;</li><li>in addition, a unique IDs shall be determined for the aggregate flows with a certain method, e.g., Diffserv Code Point in IPv4/v6, flow-label in IPv6, EXP in MPLS, or 802.1p in Ethernet etc., and mapping to each other is required across heterogeneous networks;</li><li>the resource allocation method may be implemented through signaling or policy server;</li><li>the requirements for the signaling protocol in a QOS domain include: <ol id="ol0002" compact="compact"><li>1. it shall initiate from and end at QERs;</li><li>2. it shall run on the QCRs along the path;</li><li>3. it shall have request, allocation, cancel, acknowledgement, and error report messages;</li><li>4. the data carried in the request message shall include source QER, destination QER, aggregate flow ID, and traffic parameter description, etc;</li><li>5. the data carried in the allocation message shall include: source QER, destination QER, aggregate flow ID, acknowledged or modified traffic parameter description, etc.;</li><li>6. it shall support incremental allocation, i.e., add or reduce resources for a certain aggregate flow in a certain granularity on the basis of originally allocated resources;</li><li>7. it shall support resource preemption, i.e., an aggregate flow with high priority can preempt unused resources allocated to an aggregate flow with low priority upon request;</li><li>8. it shall guarantee reliable transmission based on the protocol on transport layer;</li><li>9. it shall support multicasting;</li><li>10. every node along the path shall maintain the state of each aggregate flow.</li></ol></li></ul>

0039The requirements for policy server include: <ol id="ol0003" compact="compact"><li>1. it can obtain link state (i.e., topology information of the QOS domain) from network devices of the QOS domain as well as information on network resource utilization and source idlesse in the QOS domain;</li><li>2. it can calculate edge-to-edge paths that meet the requirements of resource requests;</li><li>3. it can configure the quantity of reserved resources at every node along the path.</li></ol>

0040Resource allocation requires support from network nodes (network devices along the transmission path); if the network devices have enough capability, they can reserve resources for every <source QER, destination QER, and aggregate flow ID> triplet, and perform classification and dispatching on the basis of the triplets; if the devices haven't enough capability, they may reserve resources for every aggregate flow ID and perform classification and dispatching on the basis of the aggregate flows; in either of above cases, each network node has to maintain <source QER, destination QER, aggregate flow ID, reserved resources> information.

0041Resource allocation among QOS domains: after the resources are allocated and reserved for aggregate flows in QOS domains, an edge-to-edge channel is formed; however, to achieve end-to-end resource reservation, the edge-to-edge channels in multiple domains have to be connected; therefore, resources have to be allocated among QOS domains; through combining resource allocation in QOS domains and resource allocation among QOS domains, an edge-to-edge resource reservation channel can be established across the entire network; similarly, the resource allocation among QOS domains may be implemented on the basis of signaling or policy server; the requirements for the signaling protocol among domains include: <ol id="ol0004" compact="compact"><li>1. it only runs between QERs and is unnecessary to run between QCRs or between QCR and QER;</li><li>2. it is unnecessary to reserve resources if the QER belongs to two QOS domains at the same time;</li><li>3. it shall have request, allocation, cancel, acknowledgement, and error report messages;</li><li>4. the data carried in the request message shall include source QER, destination QER, aggregate flow ID, and traffic parameter description, etc;</li><li>5. the data carried in the allocation message shall include: source QER, destination QER, aggregate flow ID, acknowledgeed or modified traffic parameter description, etc.;</li><li>6. it shall support incremental allocation, i.e., add or reduce resources for a certain aggregate flow in a certain granularity on the basis of originally allocated resources;</li><li>7. it shall support resource preemption, i.e., an aggregate flow with high priority can preempt unused resources allocated to an aggregate flow with low priority upon request;</li><li>8. it shall guarantee reliable transmission based on the protocol on transport layer;</li><li>9. it shall support multicasting;</li><li>10. the QERs shall maintain the state of aggregate flows.</li></ol>

0042The requirements for policy server include: <ol id="ol0005" compact="compact"><li>1. each QOS domain shall have its own policy server; it is unnecessary that the entire network should be managed by the same policy server;</li><li>2. the policy servers shall exchange edge-to-edge resource reservation and resource idlesse information in their respective domains with each other;</li><li>3. it can obtain resource utilization information between QERs in two domains;</li><li>4. after the policy server in a QOS domain calculates the edge-to-edge resource path in the QOS domain, it informs the path to the policy server in the next QOS domain;</li><li>5. the policy server in each domain is responsible for configuring the nodes in the domain along the edge-to-edge path across the entire network;</li></ol>

0043If the QOS domains have enough capability, they may reserve resources for every (source QER, destination QER, aggregate flow ID) triplet and perform classification and dispatching on the basis of the triplets; if the QOS domain haven't enough capability, they may reserve resources for every aggregate flow ID and perform classification and dispatching on the basis of the aggregate flows.

0044Resource allocation between the application terminals and the network: after the edge-to-edge resource reservation channel across the entire network is established, it is necessary to perform resource allocation between the application terminals and the network and adjust edge-to-edge resource reservation across the network, in order to finally implement end-to-end resource allocation for data flows;

0045There are two approaches for resource allocation between the application terminals and the network: one is to allocate resources for aggregate flows; the other is to allocate resources for data flows; the second approach is a step of the first approach, i.e., pre-allocation;

0046The first approach, i.e., allocate resources for aggregate flows, is similar to resource allocation among QOS domains; since the allocated resources are for aggregate flows, the destination address of remote application is unknown, and therefore the reservation is not end-to-end; there are two approaches for resource allocation for aggregate flows: one is that the QAER allocates resources actively; the other is that the application terminals request for resources to QAERs via the policy servers; in case that the QAER host allocates resources actively, the host don't participate in resource allocation, and the resources for aggregate flows are only allocated on the QAER, which may be regarded as a proxy; in case that source allocation is performed through policy server, the host may request for resources to the policy server with out-of-band, in-band mechanisms, and, after the request is accepted, the policy server informs the QAER to allocate resources;

0047The second approach is to allocate resources for data flows and it is a part of end-to-end resource allocation; since the application terminals are diverse, the life cycles of data flows are different; therefore, the resource allocation is performed through a dynamic end-to-end signaling protocol; the requirements for the end-to-end signaling include: <ol id="ol0006" compact="compact"><li>1. it can run between an application terminal and the QAER as well as between application terminals;</li><li>2. QERs shall be aware of the protocol and process accordingly; however, messages shall be transmitted between QERs transparently through the end-to-end signaling protocol;</li><li>3. it shall have request, allocation, cancel, acknowledgement, and error report messages;</li><li>4. the data carried in the request message shall include: source address, destination address, flow ID, proposed aggregate flow ID, traffic parameter description, and list of passed QERs, etc.;</li><li>5. the data carried in the resource allocation message shall include: source address, destination address, flow ID, allocated aggregate flow ID, acknowledged or modified traffic parameter description, and end-to-end QER list, etc.;</li><li>6. as the request message is transmitted in the network and, the passed node shall identify whether it is a QER itself; if so, the node adds the its ID in the QER list in the message;</li><li>7. the QER list shall be removed from the request/allocation message between the QER and the application terminal;</li><li>8. when the end-to-end allocation message is returned to a QER, the QER shall check whether enough resources have been allocated for the aggregate flows between itself and another QER; if not, the signaling protocol mechanism in the domain/between the domains is triggered to perform incremental resource allocation between the two QERs;</li><li>9. the QER connected to the source shall determine the class of the aggregate flow; it may use the proposed class in the request message or assign a different class;</li><li>10. the request message shall be retransmitted periodically to detect whether the QERs along the end-to-end path have changed;</li><li>11. it shall guarantee reliable transmission with a protocol on transport layer;</li><li>12. it shall support multicasting;</li><li>13. the protocol used may be the same as the protocol for resource allocation between domains; however, the working between application and QER or between QERs may be a different one;</li><li>14. the resource allocation protocol in the domain can be distinguished easily;</li><li>15. only the QAER maintains the states of data flows; the QAER connected with the destination shall contain a QER list; other QERs don't maintain the states of data flows.</li></ol>

0048The resource allocation process for data flows as described above is an end-to-end resource allocation process; whereas the resource allocation process for aggregate flows comprises resource allocation between application terminal and QAER, resource allocation in QOS domains, and resource allocation between QOS domains and are performed in individual sections; the combination of the two resource allocation mechanisms constitute the QOS resource allocation process across the entire network.

0049To simplify the implementation, the end-to-end signaling protocol may be the same as that used between QOS domains, i.e., messages for data flow and aggregate flow are transmitted through the same protocol, however, in different application environments, the protocol behaves as the resource allocation protocol for data flows or resource allocation protocol for aggregate flows.

0050Hereunder, the method according to the present invention (i.e., the entire process of end-to-end resource allocation) is described with reference to the attached drawings:

0051As shown in <figref idref="f0002">Fig.4</figref>, APP1 is the source terminal, APP2 is the destination terminal, R1 - R4 are four QOA edge routers, PS1 is the policy server of QOS domain 1 (QOS-AS1), and PS2 is the policy server of QOS domain 2 (QOS-AS2); the resource allocation process in the network structure in <figref idref="f0002">Fig.4</figref> is shown in <figref idref="f0003">Fig.5</figref>, comprising: <ul id="ul0006" list-style="none" compact="compact"><li>step 1: the source terminal APP1 initiating a resource request message containing source address <APP1>, destination address <APP2>, flow ID <Source Port, Dest Port, Protocol>, proposed aggregate flow ID <DSCP1>, and traffic parameter description, etc.;</li><li>step 2: the edge router R1 directly connected with source terminal APP1 interacting with the policy server PS1 to judge whether the request can be accepted; if the request is rejected, returning a request failure message; if the request is accepted, determining that the aggregate flow ID is DSCP2, which may be identical to or different from DSCP1; then checking whether there are enough resources for aggregate flow DSCP2 between the source terminal APP1 and the edge router R1; if the resources are not enough, returning a request failure message; otherwise adding R1 ID in the QER list of the resource request message and forwarding the resource request message;</li><li>step 3: when the resource request message passes through the edge router R2, R2 confirming itself is the QER of the data flow through detecting that the interface at the previous hop or the next hop belongs to different QOS-ASs, adding ID of itself in the QER list of the request message, and forwarding the resource request message;</li><li>step 4: at the edge router R3, since the message enters into a new QOS domain, the edge router R3 interacting with the policy server PS2 of QOS-AS2 to judge whether to accept the resource request; if the request is rejected, R3 returning a resource request failure message; otherwise R3 forwarding the request message;</li><li>step 5: at the edge router R4, which is directly connected with the destination terminal APP2, confirming itself is the QAER for the data flow through detecting that the interface at the next hop is connected with the destination terminal, adding ID of itself in the QER list of the request message, and judging whether there are enough resources for the aggregate flow between the edge router R4 and the destination terminal APP2; if the resources are not enough, returning a request failure message to the source terminal APP1; otherwise storing the QER list, and forwarding the request message with the removed QER list to the destination terminal APP2;</li><li>step 6: the destination terminal APP2 determining the available resources for data flow and returning an allocation message to the source terminal APP1;</li><li>steps 7 and 8: at the edge router R4, allocating resources for the aggregate flow between the edge router R4 and the destination terminal APP2, adding the QER list in and forwarding the resource allocation message, and judging whether there are enough resources between the edge router R4 and the edge router R3 at the same time; if the resources are enough, transmitting the resource allocation message for the data flow to the edge router R3; otherwise, at the edge router R4, activating the resource allocation process for the aggregate flow between the edge router R4 and the edge router R3, and at the same time transmitting the resource allocation message for the data flow and a resource allocation message for the aggregate flow together to the edge router R3, wherein in an IP network, the two messages can be included in the same protocol packet, and when the messages reaches to the edge router R3, the edge router R3 retrieves the R4 ID from the QER list; if a node between the edge router R4 and the edge router R3 can't meet the resource allocation requirements, it sending an allocation failure message for the aggregate flow to the edge router R4 as well as a request failure message for the aggregate flow to the edge router R3; when receiving the failure messages, the edge router R4 sending an end-to-end allocation failure message to the destination terminal APP2;</li><li>steps 9, 10 and 11: when the allocation message passes through the edge router R3, the edge router R2, or the edge router R1, each edge router (R3, R2, or R1) retrieving the previous QER ID from the QER list, checking whether there are enough reserved resources between itself and the previous QER or the next QER; if the resources are enough, the QER forwarding the allocation message to the source terminal APP1, otherwise activating the signaling protocol process in the domain, between the domains to allocate resources; if the allocation is successful, the QER continuing to forward the end-to-end resource allocation message to the source terminal APP1; if the allocation fails, the QER sending a request failure message to the source terminal APP1 and sending an allocation failure message to the destination terminal APP2;</li><li>if the QER in a QOS domain doesn't record the state of resource reservation for aggregate flow between itself and the previous QER in the transmission direction of the data flow, the method used for resource adjustment for the aggregate flow shall be slightly different from that described in steps 7 and 8; please refer to steps 12, 13 and 14, the method shall be: the edge router R1 checking whether there are enough resources between itself and the edge router R2; if the resources are enough, removing the QER list and sending the end-to-end resource allocation message to the source terminal APP1; if the resources are not enough, the edge router R1 sending a resource request message for the aggregate flow to the edge router R2; when receiving the resource allocation message from the edge router R2, removing the QER list and sending the end-to-end allocation message to the source terminal APP1, such a resource allocation mode supports multicast resource allocation and can speed up establishment of the resource reservation path;</li><li>step 15: the source terminal APP1 allocating resources for the data flows and accomplishing the end-to-end resource allocation process.</li></ul>

0052Hereunder the working procedures of the policy server in the method according to the present invention are described in detail with reference to the attached drawings. As shown in <figref idref="f0004">Fig. 6</figref>, there is a policy server in each of the QOS-Ass, which is referred to as PS1 and PS2 respectively; the policy server PS1 and the policy server PS2 exchange information about their QOS-ASs and the managed edge nodes with each other and collect information on resource allocation at the nodes in QOS-ASs; in addition, the policy server PS1 and the policy server PS2 allocate resources for aggregate flows in domains and between the QERs and the application terminals and cooperate with each other to allocate resources for the aggregate flows between the QOS domains; the specific procedures include: <ol id="ol0007" compact="compact"><li>1. the source terminal APP1 initiating a resource request to the policy server PS1, with the destination address as destination terminal APP2; said resource request message containing source address <APP1>, destination address <APP2>, flow ID <Source Port, Dest Port, Protocol>, proposed aggregate flow ID <DSCP1>, and traffic parameter description, etc.;</li><li>2. the policy server PS1 determining the aggregate flow ID as DSCP2, which may be identical to or different from DSCP1, learning that the path will enter into QOS-AS2 through searching in the route table, etc., of the nodes in the domain, and forwarding the request message to the policy server PS2;</li><li>3. the policy server PS2 learning that the destination terminal APP2 is directly connected with the current domain through searching in the route table, etc., of the nodes in the domain, and forwarding the request message to the destination terminal APP2;</li><li>4. at the destination terminal APP2, determining the resource that is available to the data flow and returning the resource allocation message to the policy server PS2;</li><li>5. the policy server PS2 checking whether there are enough resources between the edge router R3 and the edge router R4; if there are not enough resources, further checking whether there are enough remaining resources; if the remaining resources are not enough, sending a request failure message to the policy server PS1 and sending a resource allocation failure message to the destination terminal APP2; if the remaining resources are enough, allocating resources for the aggregate flows between the edge router R3 and the edge router R4 in an incremental manner, and forwarding the allocation message to the policy server PS1;</li><li>6. processing at the policy server PS1 being the same as that at the policy server PS2;</li><li>7. allocating resources at the source terminal APP1.</li></ol>

0053In the case of a single IP QOS domain, the simplified processing procedures of the method according to the present invention are as follows: <ul id="ul0007" list-style="none" compact="compact"><li>in the single IP QOS domain as shown in <figref idref="f0004">Fig.7</figref>, above procedures can be simplified as follows: the end-to-end signaling protocol and the protocol for resource allocation in the domain are combined into one protocol, and it is unnecessary to record the QER list; the specific working procedures comprise: <ol id="ol0008" compact="compact"><li>1. the source terminal APP3 initiating a resource request message containing source address <APP3>, destination address <APP4>, flow ID <Source Port, Dest Port, Protocol>, proposed aggregate flow ID <DSCP3>, and traffic parameter description, etc.;</li><li>2. determining the aggregate flow ID as DSCP4 at the edge router R5 directly connected with source terminal APP3, which may be identical to or different from DSCP3;</li><li>3. transmitting the end-to-end request at the nodes between the edge router R5 and the edge router R6 transparently;</li><li>4. at the edge router R6, recording the data flow ID and corresponding aggregate flow ID;</li><li>5. at the destination terminal APP4, determining the available resources for the flow, returning an allocation message to the source terminal APP3; said message containing application address <APP3>, destination address <APP4>, flow ID <Source Port, Dest Port, Protocol>, allocated aggregate flow ID <DSCP4>, and acknowledged traffic parameter description, etc.;</li><li>6. allocating resources for the aggregate flow on the entire path between the edge router R6 and the edge router R5; if the resources are not enough, returning a resource request failure message to the source terminal APP3 and sending an allocation failure message to the destination terminal APP4;</li><li>7. at the edge router R5, recording the data flow ID and the corresponding aggregate flow ID;</li><li>8. at the source terminal APP3, allocating resources for the data flow.</li></ol></li></ul>

0054The method according to the present invention further comprises the processing procedures in case that the end-to-end path changes: <ul id="ul0008" list-style="none" compact="compact"><li>when the end-to-end path changes, the control plane shall be aware of such change and change the resource allocation accordingly.</li></ul>

0055In signaling protocol processing case, as shown in <figref idref="f0002">Fig.4</figref>, the processing procedures comprise: <ol id="ol0009" compact="compact"><li>1. storing the QER list information in the state of data flow on the QER, i.e., the edge router R4, that is directly connected with the destination terminal APP2;</li><li>2. the source terminal APP1 retransmitting the end-to-end request message periodically to detect whether the path has changed;</li><li>3. when the request message reaches to the QAER, i.e., the edge router R4, that is directly connected with the destination terminal APP2, there being three cases: 1) the current QAER is different from the previous passed QAER; herein the current QAER is referred to as the edge router R4'; 2) the current QAER is the same as the previous passed QAER, but the QER lists thereof are different; 3) the QAERs are the same, and so are the QER lists; then going to steps 4, 5 and 6, respectively;</li><li>4. in the case of above 1), the new QAER, i.e., the edge router R4' returning a new end-to-end allocation message; the old QAER, i.e., the edge router R4, as the result of timeout, sending a Cancel Allocation message to the source terminal APP1 and sending a Cancel Request message to the destination terminal APP2;</li><li>5. in the case of above 2), the edge router R4 creating new flow state information and sending a new allocation message (containing the new QER list) to the source terminal APP1, and, after the flow state information (containing the old QER list) times out, sending a Cancel Allocation message to the source terminal APP1;</li><li>6. in the case of above 3), no processing.</li></ol>

0056In the case of policy server processing, the processing procedures are as follows: <ol id="ol0010" compact="compact"><li>1. storing the list of passed QERs in the current QOS domain in the data flow state at the policy server PS1 and the policy server PS2;</li><li>2. the source terminal APP1 retransmitting the end-to-end request message periodically to detect whether the path has changed;</li><li>3. when the request message reaches to the policy server PS2 via the policy server PS1, the policy server PS1 and the policy server PS2 calculating whether the passed QERs in the corresponding domains have changed, respectively; if the QERs have changed, canceling the resource allocation for the aggregate flow between the old QERs, and allocating resources for the aggregate flow between the new QERs;</li><li>4. the policy server PS1 and the policy server PS2 cooperating with each other to calculate whether the pair of QERs across domainss has changed; if the pair of QERs has changed, canceling the resource allocation for the aggregate flow between the old QERs, and allocating resources for the aggregate flow between the new QERs.</li></ol>

0057The method according to the present invention further comprises processing procedures of resource dispatching for the aggregate flow; said processing procedures comprising the following three procedures: <ul id="ul0009" list-style="none" compact="compact"><li>statistical adjustment of pre-allocated resources: during initialization, edge-to-edge resources are pre-allocated, so that frequent signaling interaction and resource allocation and release can be avoided in most cases; however, if the pre-allocated resources don't meet the statistical characteristics of the aggregate flow, such a mechanism will not provide expected efficacy; thus adjustment of pre-allocated resources has to be performed periodically or aperiodically; for end-to-end signaling, the time interval between such adjustments is longer;</li><li>the adjustments may be performed manually or automatically, whenever the resources are found as insufficient or excessive for a long time; it can be seen that the basis of such adjustments is measurement of usage of edge-to-edge resources, as follows: <ol id="ol0011" compact="compact"><li>1. recording each pair of edge-to-edge pre-allocated resources at the QER;</li><li>2. setting a measuring interval;</li><li>3. calculating the edge-to-edge mean rate, time delay, etc., within the interval;</li><li>4. adjusting the resources within the next internal to meet the statistical characteristics within the next internal.</li></ol></li></ul>

0058Usually, the subscriber's participation is required when the bandwidth is adjusted between a host and the QAER, for example, when the subscriber subscribes some kind of service in an in-band or out-of-band manner, the network allocates appropriate bandwidth to the subscriber.

0059Incremental adjustment: as the result of aggregation, edge-to-edge resource allocation is incremental, that is to say, adding or reducing resources based on originally allocated resources, instead of updating resources in an end-to-end manner just like the end-to-end. The two signaling protocols shall be different from each other in the request and allocation messages so as to enable the messages to differ from the messages for resource reservation; to further reduce frequent resource dispatching, resources may be allocated in an incremental manner at a certain granularity (called resource granularity); each resource granularity determines a fixed quantity of resources, e.g., each granularity can be determined as 100K; for example, if the requested quantity of resources is 350K, 4 granularities may be allocated to meet the current resource request; at the same time, it can meet subsequent resource requests to certain extent, i.e., the quantity of allocated resources for the current aggregate flow is greater than or equal to the requested quantity of resources; during resource release, if the quantity of resources to be released hasn't reached to the boundary of the granularity, no resources will be released for the aggregate flow, until the accumulative quantity of resources to be released reaches to the boundary of the granularity.

0060Resource preemption: though edge-to-edge resources are pre-allocated for aggregate flows, an aggregate flow may preempt resources pre-allocated to an aggregate flow with low priority when it has a new resource request but the pre-allocated resources are not enough; the sequence is: from the lowest priority up to the current priority; when the preempted resources are released, starting from the aggregate flow with the next priority, it will judge whether the current resources of the aggregate flow are less than the pre-allocated resources, and release the resources to it if so.

0061The unit of resource preemption is also resource granularity. For a QOS domain with only one router and interfaces thereof, the above three processing procedures can also be performed at a single router.

0062In the present invention, when requesting for resources between the application terminal and the network, the request message carries the flow ID, through which the QAER can therefore identify the flow and set priority for it according to the allocated aggregate flow ID; the QAER also determines the Commit Access Rate (CAR) to ensure the data flow meets the stated traffic characteristics.

0063In the present invention, local resource dispatching shall take advantage of IP statistical multiplexing; the reserved resources of the aggregate flows can be shared dynamically among the aggregate flows; in a processing cycle, unused resources of aggregate flows with high priorities can be used by aggregate flows with low priorities.

0064In the present invention, to ensure the data flow path equals to the aggregate flow path (i.e., ensure the data flow is transmitted along the reserved path), the following solution may be used: in a QOS-AS, the path of a data flow is defined with source address and destination address of the flow, while the path of an aggregate flow is defined with addresses of the QERs at two ends; in special cases, it is possible that the two paths don't equal to each other, for instance: <ol id="ol0012" compact="compact"><li>1. Route policy or policy-based routing is enforced at a node along the path, causing the route is not the Metric-based optimal route; such a node is usually an ASBR or ABR;</li><li>2. There are multiple equivalent paths in the QOS-AS but resource allocation is only carried out on one of them; resources are not allocated on other paths;</li></ol>

0065In the first case, if QER equals to ASBR/ABR and there is no node on which route policy/policy-based routing is enforced in the QOS-AS, the two paths equal to each other; if the path of the aggregate flow is different to that of the data flow passing through the QOS-AS, the issue can be solved through establishing a tunnel between the QERs and forcing to forward the application flows through the tunnel;

0066In the second case, the issue can be only solved through establishing a tunnel between the QERs.

0067During the working process, the end-to-end signaling protocol involved in the present invention shall identify the QERs for the flow as follows: <ul id="ul0010" list-style="none" compact="compact"><li>At the network device, configuring the interfaces of it into 3 types: <ol id="ol0013" compact="compact"><li>a. Subscriber interface, which is connected to the subscriber directly or via a network without QOS;</li><li>b. Interface in QOS-AS, which is directly connected to a node in the QOS-AS; (default)</li><li>c. Interface between QOS-ASs, which is connected to a node in a different QOS-AS.</li></ol></li><li>If a QER is cross two QOS-ASs, all of its interfaces shall be configured as interfaces between QOS-ASs, and the QOS-AS IDs shall be configured for it.</li></ul>

0068When a request message reaches to the network device, it calculates the interface at previous hop/next hop, and determines the QER according to the combination of interfaces at next hop/previous hop as follows: <ul id="ul0011" list-style="none"><li>a/a: Connected to subscribers at both ends, QAER;</li><li>a/b or b/a: QAER;</li><li>a/c or c/a: QAER;</li><li>b/b: not QER;</li><li>b/c or c/b: QER;</li><li>c/c: Judging whether the QOS-AS IDs of the two interfaces are the same; if they are not the same, it is a QER; otherwise it is not a QER.</li></ul>