Results 1 
3 of
3
Mapping Neural Networks onto MessagePassing Multicomputers
, 1989
"... This paper investigates the architectural requirements for simulating neural networks using massively parallel multiprocessors. First, we model the connectivity patterns in large neural networks. A distributed processor/memory organization is developed for efficiently simulating asynchronous, value ..."
Abstract

Cited by 23 (13 self)
 Add to MetaCart
This paper investigates the architectural requirements for simulating neural networks using massively parallel multiprocessors. First, we model the connectivity patterns in large neural networks. A distributed processor/memory organization is developed for efficiently simulating asynchronous, valuepassing connectionist models. Based on the network connectivity and mapping policy, we estimate the volume of messages that need to be exchanged among physical processors for simulating the weighted connections of a neural network. This helps determine the interprocessor communication bandwidth required, and the optimal number and granularity of processors needed to meet a particular cost/performance goal. The suitability of existing computers is assessed in the light of estimated architectural demands. The structural model offers an efficient methodology for mapping virtual neural networks onto a real parallel computer. It makes possible the execution of largescale neural networks on a mod...
StorageEfficient, DeadlockFree Packet Routing Algorithms for Torus Networks
 IEEE Trans. on Computers
, 1994
"... We present two new packet routing algorithms for parallel computers with torus interconnection networks of arbitrary size and dimension. Both algorithms use only minimal length paths, are fully adaptive in the sense that all minimal length paths may be used to avoid congestion, and are free of deadl ..."
Abstract

Cited by 8 (1 self)
 Add to MetaCart
(Show Context)
We present two new packet routing algorithms for parallel computers with torus interconnection networks of arbitrary size and dimension. Both algorithms use only minimal length paths, are fully adaptive in the sense that all minimal length paths may be used to avoid congestion, and are free of deadlock, livelock and starvation. Algorithm 1 requires only three central queues per routing node. It is the first known minimal length packet routing algorithm for torus networks which requires a constant number of queues per node, regardless of the size and dimension of the torus. In fact, the requirement of three queues per node is optimal, as no such algorithm is possible when all nodes have two or fewer queues. Algorithm 2 requires only that each node have two input buffers per edge. It is the first known minimalfullyadaptive packet routing algorithm for torus networks which does not require central queues and which does not require any node to have more than two input or two output buffe...
VIRTUAL CHANNELS IN WORMHOLE ROUTERS
, 1999
"... This paper analyzes the impact of virtual channels on the performance of wormhole routing algorithms. We study wormhole routing on network in which each physical channel, i.e., communication link, can support up to B virtual channels. We show that it is possible to route any set of messages with L f ..."
Abstract
 Add to MetaCart
(Show Context)
This paper analyzes the impact of virtual channels on the performance of wormhole routing algorithms. We study wormhole routing on network in which each physical channel, i.e., communication link, can support up to B virtual channels. We show that it is possible to route any set of messages with L flits each, whose paths have congestion C and dilation D in O((L+ D) C(D log D) 1 B B) flit steps, where a flit step is the time taken to transmit B flits, i.e., one flit per virtual channel, across a physical channel. We also prove a nearly matching lower bound; i.e., for any values of C, D, B, and L, where C, D B+1 and L=(1+0(1)) D, we show how to construct a network and a set of Lflit messages whose paths have congestion C and dilation D that require 0(LCD 1 B B) flit steps to route. These upper and lower bounds imply that increasing the buffering capacity and the bandwidth of each physical channel by a factor of B can speed up a wormhole routing algorithm by a superlinear factor, i.e., a factor significantly larger than B. We also present a simple randomized wormhole routing algorithm for the butterfly network. The algorithm routes any qrelation on the inputs and outputs