http:^^millennium.cs.ucla.edu:80^~ssn^
来自「This data set contains WWW-pages collect」· EDU:80^~SSN^ 代码 · 共 129 行
EDU:80^~SSN^
129 行
Date: Wednesday, 20-Nov-96 19:28:39 GMTServer: NCSA/1.3MIME-version: 1.0Content-type: text/htmlLast-modified: Saturday, 27-May-95 00:04:02 GMTContent-length: 5559<html> <head><title>The Distributed Supercomputer Supernet (SSN)</title></head><HEADER><H1>The Distributed Supercomputer Supernet (SSN)</H1></HEADER><BODY><DL><DT>ORGANIZATION: <!WA0><a href="http://www.ucla.edu">University of California, Los Angeles </a><DT>SUBCONTRACTORS: Jet Propulsion Laboratories<DT>PRINCIPAL INVESTIGATORS: <!WA1><a href="http://millennium.cs.ucla.edu/lk.html">Leonard Kleinrock</a> and Larry Bergman<DT>CO-PRINCIPAL INVESTIGATORS:<DD>Nicholas Bambos<br>Jason Cong <br>Eli Gafni <br>Mario Gerla <br></DL><!WA2><img src="http://millennium.cs.ucla.edu:80/~ssn/Pictures/ssn1.gif"><hr><H2>OBJECTIVE:</H2>The major focus of this research is the design, testing and prototypingof a novel, high-performance, optical interconnection network forsupercomputers, which is based on optically communicating mesh routersor crossbars, and which is scalable. The geographic coverage rangesfrom interdepartmental to campus and even to metropolitan areas. Thenetwork provides very high speed multiple services, supporting hybridcircuit-switched and datagram traffic, and direct or multi-hopconnections that are dynamically reconfigurable. At a first networkinglevel, we locally interconnect workstations, supercomputers, peripheraldevices, mass memory etc. through host interfaces. At a highernetworking level, we use fully optical interconnects, allowingcommunication between devices connected to distinct mesh routers. Agoal of this research is to capture the large, latent, distributedcomputational power of the network processors, to be used for networkcontrol and management, leading to an intelligent network. A mainmotivation for the research stems from the limitations observed incurrent supercomputer interconnect systems, and from the opportunitiesoffered by emerging communication technologies such as WDMoptoelectronics for novel, feasible system architectures.<H2>APPROACH:</H2>We propose to overcome the usual interconnect problems by replacing thepoint-to-point links with an all-optical interconnect system. Namely,the high speed LANS will be connected to an optical star (or tree)"physical" topology. Wavelength Division Multiplexing (WDM) will beused to subdivide the very large fiber bandwidth into several channels,each of Gbps bandwidth. WDM channels (supporting also time divisionmultiplexing) will be established between modules, thus defining adense "virtual" interconnection topology, which is dynamicallyreconfigurable, responding to changing traffic patterns. A pool ofchannels will be set aside for direct, end-to-end connections betweenmesh routers, providing circuit-switched service for real-time trafficapplications.<P>The heart of the Distributed Supercomputer Supernet is OPTIMIC -OPTical Interconnect of MyriNet IC chips. OPTIMIC is a novel,high-performance network consisting of Myricom asynchronous pipelinecrossbars (APC) interconnected by a WDM optical backbone. It supportsboth circuit-switched and packet switched (datagram) traffic; itachieves virtual topology reconfigurable interconnection through anoptical star (or tree), and, it bases its networking operations on theintelligent fabric of the network itself. In OPTIMIC, each network node(i.e. APC) is connected to local hosts, to other nodes, and to multipleoptical channels via an Optical Channel Interface (OCI).The optical interconnection and the use of their intelligence(computational power) for network operation and control, will besupported by several technical innovations at various levels of networkdesign, including system integration and implementation, interfacing,distributed algorithms and protocols, modeling and performanceevaluation, intelligent control and resource allocation etc.<hr><!WA3><img src="http://millennium.cs.ucla.edu:80/~ssn/Pictures/ssn2.gif"><hr><H2>TECHNOLOGY TRANSITION:</H2>Among the technology transfer possibilities for the research, we list the following.Fine Grain Meta-Supercomputer: The SSN attributes would accelerate the evolution of a network-based operating system with precise synchronization of dispersed processes, fine grain process management on 100's-1000's of processor elements, distributed checkpointing of jobs, and dynamic entry of new hosts.<P>Real Time Distributed Network Operating System: Low and predictable(bounded) latency makes ideal for wide area network control and dataacquisition applications. Examples in the government include Air ForceSATCOM network, SDIBE, remote robot control for NASA applications, andin the commercial arena, oil refinery and power plant control, avionicsand spacecraft control systems, control of electrical powerdistribution systems, and factory automation.<P>Distributed Image Data Base Perusal: Scientific image-based data-basearchival and perusal systems are now being developed in severalefforts, such as the UC Sequoia effort and the MAGIC testbed. NASAapplications, such as EOS will require the capability of perusingthrough terabytes of data very quickly and interactively. A low latencyhigh throughput network will be essential for responding quickly tointeractive control from the user (datagram) and sending image burstsback to the user (streams/circuit switched).<hr><!WA4><img src="http://millennium.cs.ucla.edu:80/~ssn/Pictures/ssn3.gif"><hr><li> <!WA5><a href="http://millennium.cs.ucla.edu:80/~ssn/status.html"> Click here</a> for the SSN Status Report.<p><li> <!WA6><A href=http://millennium.cs.ucla.edu:80/~ssn/docs.html>Click here</A> for more SSNdocumentation. <P><hr></DL><address><!WA7><a href="http://millennium.cs.ucla.edu/~kolias">kolias@cs.ucla.edu</a></address><!--hhmts start -->Last modified Tue Mar 21 16:30:16<!-- hhmts end --></body> </html>⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?