CN100461748C - Method for interconnecting switched network route directly - Google Patents

Method for interconnecting switched network route directly Download PDF

Info

Publication number
CN100461748C
CN100461748C CN 200410085381 CN200410085381A CN100461748C CN 100461748 C CN100461748 C CN 100461748C CN 200410085381 CN200410085381 CN 200410085381 CN 200410085381 A CN200410085381 A CN 200410085381A CN 100461748 C CN100461748 C CN 100461748C
Authority
CN
Grant status
Grant
Patent type
Prior art keywords
switching
deadlock
plane
cell
buffer
Prior art date
Application number
CN 200410085381
Other languages
Chinese (zh)
Other versions
CN1764150A (en )
Inventor
李英新
凯 王
王少勇
胡致远
邹昕光
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Grant date

Links

Abstract

一种直接互连交换网路由的方法,将交换节点中的缓存队列分组,各交换节点中具有相同目的地址的缓存队列组形成虚拟输出缓存队列VOQ交换平面,信元在各VOQ交换平面中进行路由,在交换节点还配置用于解除死锁的缓存队列组,当发生死锁时,将队列头信元送到由各交换节点中用于解除死锁的缓存队列组所形成的交换平面中进行路由;用于解除死锁的缓存队列组所形成的交换平面包括两个交换平面,其中一个交换平面对发生死锁的信元进行局部恢复,一个交换平面对发生死锁的信元进行全局恢复,本发明提供了一种消除死锁的直接互连交换网路由方法,提高了交换网通道利用率,避免了交换网络丢包。 A direct method of interconnecting the switching network, the packet switching nodes in the queue buffer, the switching nodes in the buffer queue group having the same destination address form the virtual output queue VOQ buffer switching planes, each cell in the VOQ switching plane routing, switching node further configured to buffer queue group released deadlock when the deadlock occurs, the queue header of the cell to the switch plane buffer queue group for each switching node releasing the formed deadlocks routing; switching plane for releasing the deadlock buffer queue group formed include two switching planes, a switching plane on which the cell recover local deadlock occurs, a switching of the plane of the cell deadlock globally recovery, the present invention provides a direct interconnection switching network for eliminating deadlock by the process, to improve the channel utilization of the switching network, switching network packet loss is avoided.

Description

直接互连交换网路由的方法 The method of switching network by the direct interconnection of

技术领域 FIELD

本发明涉及通信技术领域,尤其涉及一种在直接互连拓朴结构的交换网中进行路由的方法。 The present invention relates to communications technologies, and in particular relates to a method of routing in a switching network of direct interconnection topology.

背景技术 Background technique

随着通信网络规模的不断扩大,要求交换网(Switching fabric)能够实现经济扩展和提供服务质量(QoS)保障。 With the continuous expansion of the scale of the communication network, requiring switching network (Switching fabric) to achieve economic expansion and quality of service (QoS) guarantees. 传统的(动态)间接互连结构交换网,如Crossbar (交换矩阵)、三级Clos等,采用集中交换背板结构,扩容时代价过大。 Traditional (dynamic) indirect switching network interconnection structure, such as on Crossbar (switching matrix), and other three-stage Clos, centralized switching backplane structure, when the cost of expansion is too large. 而直接互连拓朴结构网络,如三维Torus(圆环),可以提供线性扩展结构,实现平滑扩容,且这种多路径网络结构提供了良好的容错性。 Direct interconnection network topologies such as 3D Torus (ring) may be provided linearly expandable structure, smooth expansion, and this multi-path network architecture provides good fault tolerance. 近十多年来,直接互连网络一直用于并行计算机结构中,有很高的可扩展性、经济性、鲁棒性,然而直接互连的计算机网络未能提供路由器所需的QoS保障,其"树饱和(tree-saturation)"的阻塞特性,不能保证包时延。 Over the past decade, has direct interconnection network for parallel computer architecture, is highly scalable, economical, robust, however, direct interconnection of computer networks can not provide QoS guarantees required router, its "saturated tree (tree-saturation)" blocking properties, packet delay can not be guaranteed.

"树饱和(tree saturation)"是因节点緩存竟争而产生的网络阻塞, 网络中通往"热点端口"路径上"中间节点"的緩存资源可能耗尽,从而阻止需要经过这些"中间节点,,到其他端口的业务流通过,进而诱发更多的网络緩存资源耗尽、直至全网瘫痪。目前解决树饱和的办法是: 交换节点配置虚拟输出緩存队列(VOQ),网络中各交换节点按照目的地址相同原则,组成不同的虚拟输出緩存队列(VOQ)交换平面(简称虚拟交换平面),以避免不同输出端口之间的緩存竟争,如图1所示,交 "Saturated tree (tree saturation)" because node caching competitive generated network congestion, network access to "hot spots port" path "intermediate nodes" cache resources may be exhausted, thereby preventing the need to go through these "intermediate nodes, , other traffic flow through the ports, and thus inducing more network cache resource exhaustion, paralysis of the whole network until the current is saturated solution tree: the virtual switching nodes arranged an output buffer queues (VOQs), the switching nodes in the network in accordance with principle the same destination address, different compositions of the virtual output buffer queues (VOQs) switching plane (referred to as a virtual switching plane), in order to avoid competition between the different cache output ports, shown in Figure 1, cross

换节点1一9中分别配置虚拟输出緩存队列(VOQ),每个节点緩存中目的地址为4的VOQ和目的地址为6的VOQ分别组成2个虚拟交换平面,目的地址为4或6的信元分别在各自的虚拟交换平面进行路由,虚拟交换平面的数目一般至少为总节点数,也可根据緩存资源适当增加。 1 in a switching node 9 are arranged virtual output buffer queue (VOQ), each node in the destination address cache and a destination address for the VOQ VOQ 4 to 6 are grouped into two virtual switching plane, the destination address is 4 or 6, the letter element in their respective virtual routing switching planes, the number of virtual switching plane is typically at least the total number of nodes may be appropriately increased according to cache resources. 采用虚拟交换平面技术解决了"树饱和(treesaturation)"带来的緩存竟争问题。 Virtual plane switching technology solves the "tree saturation (treesaturation)" competition cache problems caused.

多路径的直接互连网络的通道利用率与路由算法、流量分担算法和仲裁相关。 Channel utilization and direct interconnection network routing algorithm multiple paths, traffic and sharing algorithm Arbitration. 路由算法常分为最短路径和非最短路径两种,最短路径算法的优点是易于避免死锁;非最短路径算法的优点是有较高的通道利用率且适应流量突发。 Shortest path routing algorithms and often divided into two kinds of non-shortest path, the shortest path algorithm advantages are readily avoid deadlock; non-shortest path algorithm advantage is a higher efficiency and accommodate traffic channel burst.

常见的最短路由算法有以下三种: Common shortest routing algorithm has the following three:

1、 最小路径的源路由算法一一随机等概率选择源节点到目的节点之间的任意一条最小路径来转发分组,如图2所示; 1, the source path routing algorithm minimum probability of randomly selecting one by one the source node to the destination node between an arbitrary minimum path to forward a packet, shown in Figure 2;

2、 自适应最小路径路由算法,如图3所示; 2, adaptive minimum path routing algorithm, shown in Figure 3;

3、 维序路由算法(Dimension-Ordered ),如图4所示。 3, the dimension order routing algorithm (Dimension-Ordered), as shown in FIG. 流量分担算法的主要目的是平衡整个交换网各路径的流量负担,可 The main purpose of traffic sharing algorithm is to balance the burden on the entire switched network flow path can be

在网络源节点预先分配确定路由,或由中间交换节点动态平tf。 In the pre-allocated network node determines source routing node or by the dynamic tf intermediate switching level. 另外反馈检测信号(反压或节点故障)能够帮助各交换节点调整流量。 Further detection feedback signal (backpressure or node failure) can help adjust the flow rate of each switching node.

仲裁算法解决直接互连交换网的交换节点内部的对外通道竟争问题,公平性是其主要指标,可以采用轮循类或公平调度算法。 Foreign arbitration algorithm to solve the internal channel switching network switching nodes interconnected directly competitive issues, fairness is the main indicator can be used round-robin scheduling algorithm class or fair.

直接互连网络的多路径特性,保证了突发条件下的通过率,增加了系统容错性,而且釆用虚拟交换平面技术解决了"树饱和"带来的緩存竟争,但在非最短路由条件下,未能消除同一虚拟交换平面中节点之间可能出现的緩存循环依赖关系,易产生死锁。 Multipath feature directly interconnected network, to ensure the passing rate under the conditions of a burst, increased fault tolerance, and preclude the use of planar technology to solve the virtual switching "saturated tree" competitive brought cache, but not in the shortest route under the conditions, the same could not eliminate virtual switching cache cycle that may arise between plane node dependency, easy to produce a deadlock.

如图5所示为死锁产生机理示意图,图中产生死锁的4个交换节点各沿逆时针回环路径申请緩存资源,但不放弃其已经拥有的緩存资源。 As shown in FIG. 5 is a schematic deadlock mechanism, FIG deadlock four switching nodes of each path in a counterclockwise loop cache resources application, but do not give up its resources already has cached. 死锁现象出现在非最短路由或者因交换节点异常而釆用偏折路由策略的 Deadlock occurs in a non-shortest routing or switching nodes due to abnormal preclude the use of deflection routing policy

情况。 Happening.

目前解决死锁的主要方法有: At present, there is a method to solve deadlock:

1) 破坏资源循环依赖关系; 1) destruction of resources circular dependency;

2) l是供充分的緩存资源; 2) l is sufficient for cache resources;

3) 允许强占和丟弃。 3) allowing seizure and discarded.

上述第一种方法要限制一些路径,需釆用最短路由或其他特定路由选择规则,这样降低了通道利用率;第二种方案受到技术和成本制约; 而由于处在链路层的交换网内部不允许丟弃,从而限制了第三种方案。 The first method to limit the number of paths when the shortest route or preclude the use of other specific routing rules, which reduces the utilization of the channel; the second solution by the technical and cost constraints; in the switching network due to internal link layer not allowed to be discarded, thus limiting the third embodiment.

发明内容 SUMMARY

本发明所要解决的技术问题是:克服现有的直接互连拓朴结构交换网在采用虚拟交换平面解决了树饱和带来的緩存竟争后,仍存在非最短路由条件下易产生死锁的问题,提供一种消除死锁的直接互连交换网路由方法,并提高交换网通道利用率,避免网络丢包。 The present invention solves the technical problem: to overcome the conventional direct switched interconnection network topology in the virtual switching plane tree saturated solution brought competitive cache, there is still easy to produce non deadlock shortest route condition problems and provide a direct interconnection to eliminate deadlock by the method of switching network, and the switching network to improve the channel utilization, to avoid packet loss. 本发明为解决上述技术问题所采用的技术方案为: 这种直接互连交换网路由的方法,将交换节点中的緩存队列分组, 各交换节点中具有相同目的地址的緩存队列组形成虚拟输出緩存队列VOQ交换平面,信元在各VOQ交换平面中进行路由,其特征在于:对所述VOQ交换平面中的死锁进行消除,包括以下步骤: Aspect of the present invention to solve the above technical problem is: This direct interconnection switching network by a method, the packet switching nodes in the queue buffer, the switching nodes in the buffer queue group having the same destination address form the virtual output buffer VOQ queue switching planes, each cell in the VOQ exchange routing plane, wherein: a plane of a deadlock in the VOQ eliminate exchange, comprising the steps of:

A、 在交换节点配置用于解除死锁的緩存队列组; A, switching node configured to buffer queue group releasing deadlock;

B、 当发生死锁时,将队列头信元送到由各交换节点中的所述用于解除死锁的緩存队列组所形成的交换平面中进行路由。 B, when a deadlock occurs, the queue header of the cell to each switching node in the switching plane for releasing the deadlock buffer queue group formed routed.

所述的步骤B中,所述用于解除死锁的緩存队列组所形成的交换平面包括两个交换平面,其中一个交换平面对发生死锁的信元进行局部恢复,当该信元到达后续交换节点时,重新回到VOQ交换平面进行;洛由; 另一个交换平面则对发生死锁的信元进行全局恢复,信元在此交换平面 Said step B, the switching plane for releasing the deadlock buffer queue group formed include two switching planes, a switching plane on which the cell recover local deadlock occurs, when the cell reaches a subsequent switching node, the exchange VOQ back plane; a Rockwell; for the other switching planes cell deadlock recovery globally, in this cell switch plane

中一直路由到目的节点。 It has been routed to the destination node.

所述的步骤B中,采用队列头信元等待时间计算器进行死锁判断, 当队列头信元等待时间超过预置时间门限,则认为发生了死锁。 Said step B, the head of queue cell latency calculator deadlock determination, when the cell queue waiting time exceeding the first preset time threshold, that the deadlock occurs.

所述的预置时间门限包括下限和上限,通过预置上、下时间门限, 决定将死锁信元送到哪个交换平面进行处理,当超过下限时,送到进行局部恢复的交换平面处理;当超过上限时,送到进行全局恢复的交换平面处理。 The preset time threshold comprises an upper limit and a lower limit, by a preset upper and lower time threshold, decide which switching plane for processing the cell to the deadlock, when it exceeds the lower limit, to a process for partial recovery switch plane; when the process exceeds the upper limit switching plane, to make global recovery.

所述的解除死锁的交换平面中,交换节点中的信元优先级随等待时间延长而增加。 Releasing said switching plane in the deadlock, the cell switching node with priority waiting time extended.

所述的VOQ交换平面中的信元采用高通道利用率的非最短路由进行路由,所述用于解除死锁的緩存队列组所形成的交换平面中的信元采用无死锁路径进行路由。 Said cell switching plane VOQ high channel utilization of the non-route shortest route, the means for releasing the deadlock the cell switching plane buffer queue group formed in the deadlock-free routing path.

本发明的有益效果为:本发明通过划分直接互连拓朴结构交换网的緩存单元,分别组成"虚拟交换平面"和"死锁恢复平面",为利用不同的路由算法提高非均衡业务下的交换网通过率创造了条件;清晰的层次划分解决了路径多样性与死锁概率增大之间的矛盾;解决了直接互连拓朴结构的交换网路由算法的死锁问题,提高了交换网通过率,避免了丟包。 Advantageous effects of the present invention are: buffer unit by dividing the direct interconnection topology of the switching network of the present invention, each consisting of "virtual switching plane" and a "deadlock recovery plane" for the use of different routing algorithms to improve services under Unbalanced exchange rate Netcom had created conditions; clear hierarchical division to solve the contradiction between the path diversity and increases the probability of deadlock; solve the switched network interconnection topology directly by the deadlock algorithm improves the switching network pass rate, to avoid packet loss.

本发明采用多"虚拟交换平面"共用"死锁恢复平面",减少了解除死锁所需的緩存队列^:,并进一步细分"死锁恢复平面"为"局部,, 和"全局,,两类,分别解决了均勻业务下的死锁和突发条件的死锁对"死锁恢复平面"的缓存需求,另外,死锁检测和交换节点通道仲裁机制, 保证了交换网QoS指标。 The present invention uses multiple "virtual switching plane" common "Deadlock Recovery plane", reducing the required release the occupied buffer queue and further subdivided ^ :, "Deadlock Recovery plane" to "local,, and" Global ,, types, respectively, and to solve the deadlock deadlock conditions under burst cache requirements for uniform traffic "deadlock recovery plane", Further, deadlock detection and channel switching node arbitration mechanism to ensure that the switching network QoS criteria.

附图说明 BRIEF DESCRIPTION

图1为虚拟交换平面结构示意图; 1 is a schematic planar structure of the virtual exchange;

图2为源路由方案示意图; FIG 2 is a schematic view of a source routing scheme;

图3为自适应最短路由方案示意图; FIG 3 is a schematic diagram of the shortest route adaptive scheme;

图4为维序路由方案示意图; FIG 4 is a schematic dimension order routing scheme;

图5为死锁产生机理示意图; FIG 5 is a schematic view of the mechanism of deadlock;

图6为本发明实现方案流程图。 6 flowchart of an implementation embodiment of the present invention.

具体实施方式 detailed description

下面根据附图和实施例对本发明作进一步详细说明: The following figures and examples will be further detailed description of the invention:

直接互连拓朴结构的交换网的多路径特性,可能引发死锁,而目前常用的解决死锁的方案都有其局限性,需要寻求新的路由策略。 Multipath feature direct interconnection switching network topology, may lead to a deadlock, while currently used to solve the deadlock program has its limitations, we need to find new routing policy.

本发明的基本思想是:由于"树饱和"与"死锁"现象的原因都是要解决緩存利用问题,因此解决方案是将直接互连交换网按交换节点的ll存队列分组,形成不同的交换平面,其交换平面划分为多个"虚拟交换平面,,和"死锁恢复交换平面",两种交换平面可分别采用不同的路由算法,以避免"树饱和,,以及"死锁"现象。 The basic idea of ​​the invention is: As the "saturated Tree" with the reason "deadlock" phenomenon is the problem to be solved by the use of the cache, so the solution is directly interconnected by the switching network of switch node ll packet storage queues to form different switching plane, which plane is divided into a plurality of exchange "virtual switching planes,, and a" deadlock recovery switching planes "two switch planes may be using different routing algorithms to avoid the" tree and saturated ,, "deadlock" phenomenon .

虚拟输出緩存队列(VOQ)按照目的地址相同原则,组成不同的虚拟交换平面,虚拟交换平面的数目至少等于总节点数,可根据緩存资源适当增加虚拟交换平面,以降低死锁概率;在交换节点中利用一个额外的緩存队列组用于解除死锁,各节点中用于解除死锁的緩存队列组形成死锁恢复交换平面,如果已发生死锁,则由死锁恢复平面解决,将队列头信元送到该交换平面中进行路由。 Caching virtual output queues (VOQs) according to the same principle of the destination address, the virtual switching planes having different compositions, the number of virtual switch plane at least equal to the total number of nodes, the virtual switching plane may be increased according to the buffer resources, in order to reduce the probability of a deadlock; in the switching node in using an additional buffer queue group for releasing the deadlock, each node in the buffer queue group for releasing the deadlock deadlock recovery switching plane is formed, if a deadlock has occurred, recovery by the deadlock plane solution, the head of the queue switching the cell to the plane routing.

本发明的"死锁恢复平面"为所有的"虚拟交换平面"提供了死锁恢复机制,"虚拟交换平面"可采用高通道利用率的非最短路由算法, 采用虚拟交换平面的死锁概率较低,整个交换网使用统一的死锁恢复平面,每个死锁恢复平面对应于交换节点中的一个额外(用于解除死锁的) 緩存队列组,该死锁恢复平面使用无死锁的路由算法。 Deadlock probability "Deadlock Recovery plane" of the present invention are all non-shortest route algorithm "virtual switching plane" deadlock recovery mechanism provides "virtual switching plane" employed high channel utilization, the virtual switching plane than low, the entire switching network deadlock recovery using a unified plane, each additional deadlock plane corresponds to a recovery switching nodes (for releasing the deadlock) buffer queues, the deadlock recovery deadlock-free routing plane algorithm.

_解决死锁的基本条件是能够检测到死锁,由于直接互连网络中各交换节点的分布控制特性,难以从逻辑上判定死锁,较容易的方法是采用等待时间判断方案。 _ Basic conditions is the ability to resolve the deadlock detection of deadlock, since the distribution of the direct switching nodes interconnected network control features, is difficult to determine the deadlock logically easier approach is to use the waiting time Analyzing program. 本发明采用队列头信元等待时间计算器,当队列头信元等待时间超过预置门限,则将该队列头信元送到死锁恢复平面。 The present invention uses the first cell queue waiting time calculator, when the head of the queue waiting time the cell exceeds a preset threshold, then the queue header of the cell to the plane deadlock recovery.

"死锁恢复平面"又可分为"局部死锁恢复平面,,和"全局死锁恢复平面"。超过时间门限的信元首先送到"局部死锁恢复平面",而当其到达下一跳交换节点时,回到"虚拟交换平面"。该平面将死锁恢复功能在整个交换网中均匀分布,而对不平衡流量所产生的严重死锁将提高"全局死锁恢复平面"处理。因此,时间门限可分为下限和上限,通过预置上下时间门限,决定将死锁信元送至"局部死锁恢复平面"或"全局死锁恢复平面,,处理。 "Deadlock Recovery plane" can be divided into "local deadlock recovery ,, and plane" plane global deadlock recovery. "Exceeds the time threshold first cell to" restore local deadlock plane ", and when it reaches the next when hop switching node, return to the "virtual switching plane." the plane will deadlock recovery evenly distributed throughout the switching network, while serious imbalance deadlock flow generated will increase the "global deadlock recovery plane" process. Therefore, the time threshold can be divided into lower and upper limit, and down by the preset time threshold, the decision to the cell deadlock "local deadlock recovery plane" or "plane ,, global deadlock recovery process. 超过上限(等待时间更长)的信元将在"全局死锁恢复平面"中一直送达目的节点,中间节点不返回"虚拟交换平面"。 Exceeds the upper limit (waiting time longer) cells will always reach the destination node "global deadlock recovery plane", the intermediate node does not return a "virtual switching plane."

虚拟交换平面中,交换节点的仲裁采用轮循或其他公平调度算法; Virtual switching plane, using a round robin arbitration switching nodes or other Fair Scheduling Algorithm;

而"死锁恢复平面,,中,交换节点仲裁也可上述方案,但为了体现等待时间与信元转发紧迫性关系,可釆用动态优先级方案,其信元优先级随等待时间增加而增加,解决了交换平面内不同信元通道的竟争。 The "Deadlock recovery ,, plane, the switching node may be the above-described arbitration scheme, but the latency time in order to reflect the urgency of cell forwarding relationship, may preclude the use of a dynamic priority scheme, the priority with which a cell increases latency, solve the competitive exchange in a plane different cell channels.

如图6所示,本发明实现方案流程如下: 6, the present invention is implementation process is as follows:

1、 首先进行初始化,设置队列头信元等待时间上门限和下门限; 1, First, initialization, set the queue header of the cell and the upper threshold latency thresholds;

2、 如果信元在全局死锁恢复平面,或队列等待时间大于上门限, 则根据信元优先级在全局死锁恢复平面分配无死锁路由转发信元,或等待下次时隙处理; 2, if the cell in the global deadlock recovery plane, or queue wait time is greater than the upper threshold, the cell according to the priority assigned global deadlock recovery deadlock-free routing and forwarding plane information element, or waiting for the next processing slot;

3、 如果信元等待的时间未超过上门限而大于下门限,则将信元送到局部死锁恢复平面,根据信元优先级分配无死锁路由转发信元,或等待下次时隙处理; 3, if the cell waiting time does not exceed the upper threshold but greater than the lower threshold, the cell will be restored to a local deadlock plane, deadlock-free routing assigned forward the cell according to the cell priority, or wait for the next processing slot ;

4、 如果信元等待的时间不大于下门限,则在虚拟交换平面中进行仲裁,选择成功信元分配路由转发信元,或等待下次时隙处理。 4, if the cell is waiting time is less than the lower threshold, then the virtual exchange arbitration plane, select the cell distributing successfully forward the cell routing, processing, or waiting for the next time slot.

本发明l是供了"虚拟交换平面,,和"死锁恢复平面"组合的交换平面方案,分别解决了直接互连交换网的"树饱和,,和"死锁"问题,并提高了通道利用率,避免了网络丢包。 l for the present invention is a "virtual switching planes,, and" Recovery deadlock plane "combination scheme switching plane, respectively, to solve the direct interconnection switching network of the" tree and saturated ,, "deadlock" problem, and to improve the channel utilization, avoid packet loss.

Claims (6)

  1. 1、一种直接互连交换网路由的方法,将交换节点中的缓存队列分组,各交换节点中具有相同目的地址的缓存队列组形成虚拟输出缓存队列VOQ交换平面,信元在各VOQ交换平面中进行路由,其特征在于:对所述VOQ交换平面中的死锁进行消除,包括以下步骤: A、 在交换节点配置用于解除死锁的缓存队列组; B、 当发生死锁时,将队列头信元送到由各交换节点中的所述用于解除死锁的缓存队列组所形成的交换平面中进行路由。 1. A method for the direct interconnection of the switching network, the packet switching nodes in the queue buffer, the switching nodes in the buffer queue group having the same destination address form the virtual output queue VOQ buffer switching planes, each plane cell switching VOQ the route, wherein: a plane of a deadlock in the VOQ eliminate exchange, comprising the steps of: a, switching node configured to buffer queue group releasing deadlock; B, when the deadlock occurs, the queue header of the cell to be routed by the switching plane of the switching node a buffer queue group in a deadlock for releasing the formed.
  2. 2、 根据权利要求1所述的直接互连交换网路由的方法,其特征在于:所述的步骤B中,所述用于解除死锁的緩存队列组所形成的交换平面包括两个交换平面,其中一个交换平面对发生死锁的信元进行局部恢复,当该信元到达后续交换节点时,重新回到VOQ交换平面进行路由; 另一个交换平面则对发生死锁的信元进行全局恢复,信元在此交换平面中一直路由到目的节点。 2. The method of claim 1 by the direct interconnection switching network as claimed in claim, wherein: said step B, the switch plane for releasing the deadlock buffer queue group formed include two switching plane wherein a plane of the cell switching deadlock partial recovery, when the cell reaches a subsequent switching node, switching back plane routing VOQ; switch plane is another pair of cell deadlock recovery globally , the cell in this switching plane has been routed to the destination node.
  3. 3、 根据权利要求2所述的直接互连交换网路由的方法,其特征在于:所述的步骤B中,采用队列头信元等待时间计算器进行死锁判断, 当队列头信元等待时间超过预置时间门限,则认为发生了死锁。 3, according to claim direct interconnection switching network according to the method of claim 2, wherein: said step B, the first cell of queue waiting time Analyzing the deadlock calculator for, when the first queue wait time information element exceeds the preset time threshold, is considered a deadlock occurred.
  4. 4、 根据权利要求3所述的直接互连交换网路由的方法,其特征在于:所述的预置时间门限包括下限和上限,通过预置上、下时间门限, 决定将死锁信元送到哪个交换平面进行处理,当超过下限时,送到进行局部恢复的交换平面处理;当超过上限时,送到进行全局恢复的交换平面处理。 4, according to claim direct interconnection switching network according to the method of claim 3, wherein: said preset time threshold comprises an upper limit and a lower limit, by a preset upper and lower time threshold, the decision to send the cell deadlock when the exchange exceeds the upper limit, the overall recovery of the plane to be processed; plane to which the exchange process, when it exceeds the lower limit, to the local exchange recovery process plane.
  5. 5、 根据权利要求3所述的直接互连交换网路由的方法,其特征在于:所述的解除死锁的交换平面中,交换节点中的信元优先级随等待时间延长而增加。 5, according to claim 3, wherein the directly interconnected by the network switching method, wherein: said switching plane releasing the deadlock, the cell switching node with priority waiting time extended.
  6. 6、根据权利要求1所述的直接互连交换网路由的方法,其特征在于:所述的VOQ交换平面中的信元采用高通道利用率的非最短路由进行路由,所述用于解除死锁的緩存队列组所形成的交换平面中的信元采用无死锁路径进行路由。 6, according to claim 1, wherein the directly interconnected by the network switching method, wherein: said switching cell VOQ plane non shortest route for routing high channel utilization, the means for releasing the dead plane cell switching lock buffer queue group formed in the deadlock-free routing path.
CN 200410085381 2004-10-22 2004-10-22 Method for interconnecting switched network route directly CN100461748C (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200410085381 CN100461748C (en) 2004-10-22 2004-10-22 Method for interconnecting switched network route directly

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200410085381 CN100461748C (en) 2004-10-22 2004-10-22 Method for interconnecting switched network route directly

Publications (2)

Publication Number Publication Date
CN1764150A true CN1764150A (en) 2006-04-26
CN100461748C true CN100461748C (en) 2009-02-11

Family

ID=36748108

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200410085381 CN100461748C (en) 2004-10-22 2004-10-22 Method for interconnecting switched network route directly

Country Status (1)

Country Link
CN (1) CN100461748C (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2393015B1 (en) 2009-01-30 2015-07-01 Fujitsu Limited Information processing system, information processing device, control method for information processing device, control program for information processing device, and computer-readable recording medium
CN102137010B (en) * 2011-02-17 2013-11-06 大唐移动通信设备有限公司 Data transmitting method and communication nodes

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5805589A (en) 1993-03-04 1998-09-08 International Business Machines Corporation Central shared queue based time multiplexed packet switch with deadlock avoidance
US20020191619A1 (en) 2001-05-31 2002-12-19 Philip Shafer Network router management interface with API invoked via login stream
CN1487699A (en) 2002-09-30 2004-04-07 华为技术有限公司 Flow rate simulating method and device for exchange network test
CN1518298A (en) 2003-01-06 2004-08-04 英特尔公司 Managing method and device of VoQ
US20040199734A1 (en) 2003-04-03 2004-10-07 Oracle International Corporation Deadlock resolution through lock requeuing

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5805589A (en) 1993-03-04 1998-09-08 International Business Machines Corporation Central shared queue based time multiplexed packet switch with deadlock avoidance
US20020191619A1 (en) 2001-05-31 2002-12-19 Philip Shafer Network router management interface with API invoked via login stream
CN1487699A (en) 2002-09-30 2004-04-07 华为技术有限公司 Flow rate simulating method and device for exchange network test
CN1518298A (en) 2003-01-06 2004-08-04 英特尔公司 Managing method and device of VoQ
US20040199734A1 (en) 2003-04-03 2004-10-07 Oracle International Corporation Deadlock resolution through lock requeuing

Also Published As

Publication number Publication date Type
CN1764150A (en) 2006-04-26 application

Similar Documents

Publication Publication Date Title
Anjan et al. An efficient, fully adaptive deadlock recovery scheme: DISHA
US6424659B2 (en) Multi-layer switching apparatus and method
US6580721B1 (en) Routing and rate control in a universal transfer mode network
Aweya On the design of IP routers Part 1: Router architectures
Qiao et al. Adaptive routing in irregular networks using cut-through switches
US6741552B1 (en) Fault-tolerant, highly-scalable cell switching architecture
US7602726B1 (en) Method and system for optimizing link aggregation usage during failures
US20120033668A1 (en) IP Multicast Snooping and Routing with Multi-Chassis Link Aggregation
Glass et al. Fault-tolerant wormhole routing in meshes
US20020080790A1 (en) Universal transfer method and network with distributed switch
Mogul et al. Devoflow: Cost-effective flow management for high performance enterprise networks
US20070121499A1 (en) Method of and system for physically distributed, logically shared, and data slice-synchronized shared memory switching
US20050100035A1 (en) Adaptive source routing and packet processing
US20020049901A1 (en) System and method for implementing source based and egress based virtual networks in an interconnection network
US7187679B2 (en) Internet switch router
US20120170575A1 (en) Methods and apparatus for validation of equal cost multi path (ecmp) paths in a switch fabric system
US6987735B2 (en) System and method for enhancing the availability of routing systems through equal cost multipath
US6285679B1 (en) Methods and apparatus for event-driven routing
Wu et al. Improving routing efficiency for network-on-chip through contention-aware input selection
US20120207175A1 (en) Dynamic load balancing for port groups
US20020051427A1 (en) Switched interconnection network with increased bandwidth and port count
Duato et al. A new scalable and cost-effective congestion management strategy for lossless multistage interconnection networks
Minkenberg et al. A combined input and output queued packet switched system based on PRIZMA switch on a chip technology
US20020075871A1 (en) System and method for controlling the multicast traffic of a data packet switch
US7633940B1 (en) Load-balanced routing

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C14 Grant of patent or utility model
CF01