No abstract

No abstract

Abstract-Resource discovery is the process of locating shared resources on a computer network. Previously studied examples include efficiently finding files with a given title on a file sharing system. New developments in the application of networked computers raise the issue of dynamic resource discovery, the process of locating shared resources that are always changing. An example application is peer-to-peer computing, where a user wishes to locate idle CPU time anywhere on the network.Peer-to-peer computing is an exciting new computing paradigm. There are vast amounts of idle CPU resources scattered through the globe. We envision a peer-to-peer system to harness those resources, where every member of the network can both share their own CPU and utilize others' CPUs. In a network of hundreds of thousands of computers, resource discovery will play an important role. To avoid debilitating amounts of excess network traffic it is imperative that an efficient resource discovery algorithm be chosen. This paper's contribution to this topic is the use of gossip to reduce network traffic without sacrificing effectiveness. This project has investigated piggybacking gossip messages on other communications to increase the intelligence of searching protocols. The overhead of piggybacking the small amount of data needed is very small, and a case study by simulation shows that it can reduce network traffic by 71-84 percent.

Each node of the Intel iPSCI2 and iPSCl860 is capable of concurrent bidirectional communication, that is, each node can concurrently send and receive a message. However, link contention created by the long message communication protocol usually interferes with this source of concurrency so that half of the total capacity of the communication network is wasted. This intelference can be overcome by synchronizing communicating pairs of nodes. The cost of algorith,ms for the simple exchange, the one-to-all broadcast, the allto-all broadcast, and the complete exchange operations can be significantly reduced, compared to the cost of their usual implementations, by exploiting this approach. Performance measurements of algorithms for these communication problems support these observations.

T w o complete exchange algorithms f o r meshes are given. The modified quadrant exchange algorithm is based on the quadrant exchange algorithm and it is well suited f o r square meshes with a power of two rows and columns. The store-and-forward complete exchange algorithm i s suitable f o r meshes of' arbitrary size. A pipelined broadcast algorithm f o r meshes is also presented. This new algorithm, called ihe double hop broadcast, can broadcast long messages at slightly lower cost than the edge-disjoint fence algorithm because it uses routing trees of lower height. This shows that there is still room for improvement in the design of pipelined broadcast algorithms f o r meshes422 0-8186-5680494 $03.00 0 1994 IEEE