scispace - formally typeset
Search or ask a question

Showing papers on "Latency (engineering) published in 1992"


Journal ArticleDOI
TL;DR: A novel component performing motion estimation is presented, designed based on the three-step hierarchical search block-matching algorithm and can be applied to image communication on ISDN, MPEG, TV transmission, HDTV, etc.
Abstract: A novel component performing motion estimation is presented. This chip is designed based on the three-step hierarchical search block-matching algorithm and can be applied to image communication on ISDN (integrated services digital network) (H.261 standard), MPEG, TV transmission, HDTV (high-definition television), etc. The practical architectural design techniques and the chip features are discussed. This component has the following features: unified execution steps, low latency delay, low I/O bandwidth, regular hardware structure, and single-chip or cascaded configurations. >

17 citations


Proceedings ArticleDOI
11 Oct 1992
TL;DR: Some of the issues arising in the design of the T9000 transputer, which integrates a complete computer in a single VLSI chip of over two million transistors, are discussed.
Abstract: Some of the issues arising in the design of the T9000 transputer, which integrates a complete computer in a single VLSI chip of over two million transistors, are discussed. High performance has been achieved by extensive use of caching and a novel processor implementation. The processor uses a fat pipeline and dispatches several dependent instructions into the pipeline each cycle. The resulting processor is able to saturate a 25-MFLOP floating-point unit. The processor is supported by a communications system that supports communications at high speed and low latency. >

16 citations


Book ChapterDOI
11 Nov 1992
TL;DR: The design of a communications board that supports collective communications operations of parallel programs is described, geared towards low-end (inexpensive) parallel computer architectures.
Abstract: We describe the design of a communications board that supports collective communications operations of parallel programs. Each cluster of processors are interconnected by a bus and connected to the communications board. The communications boards are, in turn, connected to a low latency, high bandwidth, slow configuration time crossbar network. The design is geared towards low-end (inexpensive) parallel computer architectures.

1 citations


Proceedings ArticleDOI
01 Jan 1992
TL;DR: FDQ is considered, a very efficient protocol called Fair Distributed Queue (FDQ), suitable for very high speed networts over lang distances, which has lower average delay results under Poisson load than both DQ DB and DQDB with BWB, and its worst-case delay performance for messages is significantly better than D QDB with or without BWB.
Abstract: Advances in fiber optic technology now allow networks to opeozate at transmission rates greater than one gigabit per second (1Gb/s). The increase in network perfonnance promises the potential for supporting applications that require high bandwidth, low latency and predictable performance. However, the benefits to the transfer of bits node-to-node are limited. Latencies inside the nodes in the form of pocessiog delays and access delays remain the two "bottlenecks" that restrict end-to-end performance. Here, we address the second bottleneck. We consider a very efficient protocol called Fair Distributed Queue (FDQ), suitable for very high speed networts over lang distances. It uses the fiber optic medium and utilizes the full channel capacity indepmdent of the network length, the transmission speed and the number of nodes. Hence, it offers the high bandwidth required by future networks. FDQ's topology and certain other features are related to DQDB. However, it does not suffer from the asymmetry problems usociated with DQDB. The DQDB protocol has recently been accepted as the IEEE 802.6 Standard for metropolitan area networks (MANs). The purpose of this sbort paper is to summarize some of the results from, especially regarding high bandwidth implementations. We show that FDQ has lower average delay results under Poisson load than both DQDB and DQDB with BWB. Also, its worst-case delay performance for messages (assuming messages require a multiple number of slots to be transmitted) is significantly better than DQDB with or without BWB. Thus, it meets low latency requirements in addition to high bandwidth. FDQ's scalability properties allow its total length to extend over 100 Ian and transmission rates well above 1 Gb/s. This eliminates the upward mobility problems faced by current network protocols, including DQDB.