V. Karamcheti and A.A. Chien. "Do Faster Routers Imply Faster Communication?". In Proceedings of the First International Workshop on Parallel Computer Routing and Communication, pages 1 -- 15, May 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....fact, even the network delays ts are functions of P 0 [14] However, to a very good approximation, we may assume them to be constant. This is because the time to go across a network (which scales with processors) is dwarfed by the high start up times to access the network (which is a constant) [11]. 10 1e 07 1e 08 1e 09 1e 10 10 100 1000 10000 Running Time (in Clock Ticks) Number of PEs Figure 6. The dependency of running time on number of processors in a sample experiment. 1282128 matrices. The machine at hand was a MasPar MP 2. The job was run, in turn, on 16; 32; 4096 ....

V. Karamcheti and A.A. Chien. "Do Faster Routers Imply Faster Communication?". In Proceedings of the First International Workshop on Parallel Computer Routing and Communication, pages 1 -- 15, May 1994.

....(system call, buffer allocation, buffer to buffer copies, etc. This overhead traditionally accounted for a high percentage of message latency [25, 21] However, we did not considered such an overhead because some recent proposals reduce and or hide that overhead, thus exposing hardware latency [17, 2, 22, 12]. So, we only considered network hardware latency in this study. 6.1 Network Model The network is composed of a set of switches. Network topology is completely irregular and has been generated randomly. However, for the sake of simplicity, we imposed three restrictions to the topologies that can ....

V. Karamcheti and A. A. Chien, "Do faster routers imply faster communication?," in Proceedings of the Workshop on Parallel Computer Routing and Communication, May 1994.

....By considering multiple outgoing links, adaptive algorithms can increase the likelihood of cut through at intermediate nodes and balance the traffic load in the network. These algorithms, however, may also increase protocol processing overhead due to the possibility of out of order arrivals [7]. While most oblivious routing algorithms generate only minimum hop routes between the source and destination nodes, some adaptive schemes consider nonminimal routes in the hope of circumventing network congestion or faulty links. Given the unique characteristics of each routing and switching ....

V. Karamcheti and A. A. Chien, "Do faster routers imply faster communication?," in Proc. Parallel Computer Routing and Communication Workshop, pp. 1--15, June 1994.

....opportunities for adaptive routing vary depending on the network topology and the distance a packet must travel. In addition, adaptive algorithms add to the complexity of both the hardware which implements the scheme and the software which must handle the possibility of outof order arrivals [10]. Each algorithm invokes a selection function [3] which selects and orders candidate links. Network performance is greatly influenced by the interaction of this function with the communication workload. Selection functions of oblivious algorithms deterministically select a single candidate link ....

V. Karamcheti and A. A. Chien, "Do faster routers imply faster communication?," in Proc. Parallel Computer Routing and Communication Workshop, pp. 1--15, June 1994.

....due to buffer allocation at source and destination nodes, message copying between user and kernel space, packetization, in order delivery and end to end flow control. Even for a very efficient messaging layer based on active messages [20] software overhead accounts for 50 70 of the total cost [15]. Therefore, reducing the network hardware latency has a minimal impact on performance. On the other hand, messages are directly sent by the hardware in DSMs, as a consequence of remote memory accesses or coherence commands. Reducing the network hardware latency and increasing network throughput ....

V. Karamcheti and A.A. Chien, "Do faster routers imply faster communication?," in Parallel Computer Routing and Communication, K. Bolding and L. Snyder (ed.), SpringerVerlag, pp. 1--15, 1994.

.... multicomputer requires 67 sec, of This work was supported by the Spanish CICYT under Grant TIC94 0510 C02 01 which less than 1 sec is due to time on the wire [19] Even for a very efficient messaging layer based on active messages [22] software overhead accounts for 50 70 of the total cost [16]. Therefore, reducing the network hardware latency has a minimal impact on performance. Moving functionality from the software messaging layer to the routing hardware has been proposed as a way to reduce the overall communication latency [16] In particular, this work proposes implementing ....

....software overhead accounts for 50 70 of the total cost [16] Therefore, reducing the network hardware latency has a minimal impact on performance. Moving functionality from the software messaging layer to the routing hardware has been proposed as a way to reduce the overall communication latency [16]. In particular, this work proposes implementing in order delivery, end to end flow control, and fault tolerance in hardware. On the other hand, messages are directly sent by the hardware in DSMs, as a consequence of remote memory accesses or coherence commands. Reducing the network hardware ....

V. Karamcheti and A.A. Chien, "Do faster routers imply faster communication?," in Parallel Computer Routing and Communication, K. Bolding and L. Snyder (ed.), Springer-Verlag, pp. 1--15, 1994.

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