J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... packets or exponentially distributed lengths simplify analytic models, recent work shows that real multicomputer applications typically generate bimodal packet length distributions [1, 2] Recent studies have examined adapting router designs to accommodate bimodal packet length distributions [6, 7]. Message destination: Message destination distributions vary a great deal depending on the network topology and the application s mapping onto different processing elements. While many analytical and simulation studies evaluate a uniform random distribution of destination nodes, 2 this pattern ....

....and selection functions combine with a set of destination distributions to determine network performance. While other studies have looked at comparing various routing algorithms over different patterns, most studies limit the selection functions used [14 16] the range of algorithms evaluated [7, 10, 17], or the destination distributions considered [11, 18] 3 Distribution Description NodeUniform Uniform random selection of destination node MatrixTranspose Source (x; y) selects destination (y; x) BitComplement Destination node id is the bit complement of the source id BitReversal ....

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

.... packets or exponentially distributed lengths simplify analytic models, recent work shows that real multicomputer applications typically generate bimodal packet length distributions [1,2] Recent studies have examined adapting router designs to accommodate bimodal packet length distributions [6, 7]. Message destitatiot: Message destination distributions vary a great deal depending on the network topology and the application s mapping onto different processing elements. While many analytical and simulation studies evaluate a uniform random distribution of destination nodes, this pattern ....

....and selection functions combine with a set of destination distributions to determine network performance. While other studies have looked at comparing various routing algorithms over different patterns, most studies limit the selection functions used [14 16] the range of algorithms evaluated [7, 10, 17], or the destination distributions considered [11, 18] Distribution Description NodeUniform Uniform random selection of destination node MatrixTranspose Source (x, y) selects destination (y,x) BitComplement Destination node id is the bit complement of the source id BitReversal Destination ....

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on S!/stem Sciences, pp. 276 285, January 1993.

....the output links. Although each physical link services at most one virtual channel in each flit cycle, multiple virtual channels can be active at the injection and reception ports; this enables the model to represent router designs that have multiple physical or logical injection reception ports [10, 14, 15]. Upon receiving the header flits of an incoming packet, the receiver (RX) decides whether to buffer, stall, or forward the packet, based on the routing and switching policies and prevailing network conditions. By treating outbound virtual channels as individually reservable resources, the model ....

.... effective mechanism for reducing the interaction between packets while still allowing traffic to share network bandwidth [8 10, 14, 31] Exporting the virtual channel abstraction to the injection and reception ports further prevents intrusion between packets at the network entry and exit points [10, 14, 15]. By tailoring the routing, switching, and flow control policies for each virtual network, multicomputer routers can support traffic classes with conflicting performance requirements. Packets on separate virtual networks interact only to compete for access to the physical links and ports. This ....

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

....distributions [1, 2] This occurs because inter node communication often consists of large data transfers, coupled with small request and acknowledgements packets. A router design can accommodate different packet lengths by separating short and long packets onto different virtual channels [10, 11]. As a further optimization, the Segment Router Architecture [12] employs different switching schemes based on packet length; long packets use wormhole switching to limit buffer space requirements, while short packets use virtual cut through switching to reduce channel contention. Message ....

.... channels allows real time communication to coexist with best effort packets without sacrificing the performance of either class [4, 23] Similarly, the router may separate short control messages and long data packets onto different virtual channels, perhaps with different switching policies [10 12]. SPIDER can also assign routing algorithms to improve end to end performance. Although adaptive routing can reduce network latency, out of order packet arrivals can increase software processing delays at the receiving node. Opportunities for adaptive routing depend on the network topology and the ....

[Article contains additional citation context not shown here]

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

.... packets or exponentially distributed lengths simplify analytic models, recent work shows that real multicomputer applications typically generate bimodal packet length distributions [1, 2] Recent studies have examined adapting router designs to accommodate such packet length distributions [4, 5]. Message destination: Message destination distributions vary a great deal depending on the network topology and the application s mapping onto different processing elements. While many analytical and simulation studies evaluate a uniform random distribution of destination nodes, this pattern ....

....with a set of destination distributions to determine network performance. While other studies have looked at comparing various routing algorithms over different patterns, most studies limit the selection functions used, the range of algorithms evaluated, or the destination distributions considered [5, 7, 8, 11, 12]. 3 Experimental Setup In order to evaluate these algorithms, we used pp messsim [13] an object oriented discrete event simulation tool. pp mess sim supports a wide range of routing algorithms under a variety of switching schemes by decoupling them from the router models which execute them. ....

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

....algorithms to traffic patterns, especially nonuniform patterns, has been studied in many of the papers on adaptive routing. Traffic char3 acteristics such as message length, message interarrival times at the sources, and destination distributions can have significant performance implications. In [11], the first issue is studied. Many of the papers on adaptive routing address the latter two issues in the generation of non uniform traffic. Traffic patterns generated from real applications would be ideal in any evaluation, but to date, we know of only one study [12] which has examined this. ....

....256 Topology: 16 ary 2 cube Destination Distribution: Uniform, hot spot, local Message Generation Distribution: Geometric Message Length: 16 flits Switch Architecture: Inject only if number of messages of same class is below a given threshhold. Evaluation Metrics: Latency, throughput Kim [11]: Routing Algorithm: Dimension ordered, planar adaptive (minimal, partially adaptive) Virtual Channels: 1 or 2 Virtual Channel Multiplexing: Not specified Virtual Channel Allocation: Not applicable Network Size: 256 nodes Topology: 2 dimensional mesh (square ) Destination Distribution: ....

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

....of packetization is an improvement in the distribution of short message latencies. Our simulations show that under bimodal traffic loads (short and long messages) long messages not only increase the average latency of short messages, but also disperse the latency distribution significantly[10]. This dependence is shown in Fig. 2. The dispersion is reflected in increased variance of short message latencies. Although only a fraction of the small messages experience increased delay (typically less than 30 ) even a single message suffering extremely long delays will slow tightly ....

....latency of short messages. Adaptive routing allows short messages to get around blockage by selecting alternate paths. Since virtual lanes and adaptive routing both improve short message performance in very different ways, they complement each other to improve the short message latency distribution[10]. Fig. 7 shows that adaptive routing and virtual lanes can eliminate the effects of long messages on short messages almost completely. The distribution of short message latencies is improved dramatically. Comparing the advanced Nonpacketized (DOR) Nonpacketized (PAR VL) Packetized (PAR VL) 0 ....

J. H. Kim and A. A. Chien. Evaluation of wormhole routed networks under hybrid traffic loads. In Proceedings of the Hawaii International Conference on System Sciences, January 1993.

....Lanes Cost of Adaptivity 17 In this section, we characterize the cost of virtual lanes by examining router designs with from one to sixteen virtual lanes. Virtual lanes can increase channel utilization in a network by multiplexing the physical channels, allowing packets to pass one another [13, 23]. Though both virtual channels and virtual lanes use additional hardware buffers, virtual lanes require greater connectivity in the router as each virtual lane is interchangeable within its virtual channel class. Essentially, this means that the crossbars cannot be partitioned. In this section, we ....

J. H. Kim and A. A. Chien. Evaluation of wormhole routed networks under hybrid traffic loads. In Proceedings of the Hawaii International Conference on System Sciences, January 1993.

....but is omitted due to space limitations. 3 Latency is measured from the time a message is generated to the time that the tail flit of the message is delivered at the destination node. If a generated message can not be injected into 3 Some coverage of long message distributions can be found in [30]. the network immediately, then it is queued and waiting time in source queues is included in the latency. Source queue latency is included because it is germane in systems where each source may have multiple outstanding network requests. The average latencies experienced by short and long ....

J. H. Kim and A. A. Chien. Evaluation of wormhole routed networks under hybrid traffic loads. In Proceedings of the Hawaii International Conference on System Sciences, volume 1, pages 276--85, January 1993.

....amount of short and long messages. This is an interesting issue but beyond the scope of this dissertation, so it is not discussed further. Our previous work covers in detail the related issues, including the impact of packet interleaving on the average latency of messages with different sizes [96, 112, 113]. 6.3.4 Integrated services with CBR VBR ABR traffic Many existing networks and most networks in the near future need to support a wide range of applications which will require different traffic characteristics and communication services. We evaluate how effectively networks with different ....

J. H. Kim and A. A. Chien. "Evaluation of wormhole routed networks under hybrid traffic loads," in Proceedings of the Hawaii International Conference on System Sciences, Maui, HI, January 1993, vol. 1, pp. 276--85.

No context found.

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

No context found.

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

No context found.

J. H. Kim and A. A. Chien, "Evaluation of wormhole routed networks under hybrid traffic loads," in Proc. Hawaii Int'l Conf. on System Sciences, pp. 276--285, January 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC