Font Size: a A A

Message-based, error-correcting protocols for scalable high-performance networks

Posted on:2003-07-14Degree:Ph.DType:Dissertation
University:The University of New MexicoCandidate:Riesen, RolfFull Text:PDF
GTID:1468390011484813Subject:Computer Science
Abstract/Summary:
Long distance communications and the associated protocols have existed for thousands of years. Fire beacons were used in antiquity and Polybius describes various devices and protocols used to transmit messages during the Roman period. Since computers have started communicating with each other over electronic networks, the number of different protocols has exploded. Communication protocols aim to move information across a network in an efficient and reliable manner. This often requires congestion and flow control, error detection and correction, and handshaking to coordinate the information transfer.; Massively parallel supercomputers, such as the nCUBE 2, Intel Paragon, Cray T3E, and IBM SP-2, contain very fast, highly reliably networks. Similar network technology is now available in PC clusters. Examples include Myrinet and Quadrics networks. These networks, which are descendants of the supercomputer networks from the 1980's, are much more reliable, exhibit different error and degradation behavior, and are utilized in a different fashion than wide area networks commonly used in intranets and internets.; Most network communication protocols are implemented as part of the operating system (OS) on the computers wishing to communicate. With very high-speed networks, software overhead can lower the achievable performance of a network. OS bypass is a method to move data directly to and from user space, avoiding activation of the OS for data movement. This means the protocol has to be implemented on the network interface cards (NIC) which move the data, or in user space as part of the application. These NICs often have limitations, such as slow processors and limited memory, that make an efficient, scalable implementation of a communication protocol difficult.; This dissertation asserts that the differences in networks and usage warrant network communication protocols designed specifically for high-performance local area networks used in scientific computing systems. We present one such protocol, compare it against existing protocols, and show that it is more efficient and scalable than the more traditional protocols.
Keywords/Search Tags:Protocols, Networks, Scalable, Used, Communication
Related items