These days, the bandwidth and latency of interprocessor networks is the most limiting factor in reaching good speedup values on multiprocessor computers. There is, however, still potential for making better use of the...
详细信息
These days, the bandwidth and latency of interprocessor networks is the most limiting factor in reaching good speedup values on multiprocessor computers. There is, however, still potential for making better use of the available bandwidth. In this paper, we present a communication scheme which supports the data-parallel programming model and which can be totally implemented in hardware. This allows us to discard the usual software layer between application programs and the network interface, leading to an efficient bandwidth usage. The scheme could be roughly described as `ordered multicast'. Two different bus based multiprocessor systems using this scheme have been built. In order to overcome the limitation of these single board systems, the communication scheme has been adapted to a ring network. A prototype of this network was implemented and is operational at 100 MBit/s using standard optical transmitters and receivers.
Today's commercial MPP machines do not employ optics within their interconnection networks. In this paper we identify several recent trends in MPP networks and attempt to predict how these trends will affect the i...
详细信息
Today's commercial MPP machines do not employ optics within their interconnection networks. In this paper we identify several recent trends in MPP networks and attempt to predict how these trends will affect the incorporation of optical interconnect technologies in the near future. We also recognize that successful commercial systems tend to change in an evolutionary fashion, and some possible evolutionary scenarios are discussed.
Initial experimental results will be presented from the realization of a smart-pixel based general purpose parallel optoelectronic processor. A single chip array of integrated processing elements and photodetectors is...
详细信息
Initial experimental results will be presented from the realization of a smart-pixel based general purpose parallel optoelectronic processor. A single chip array of integrated processing elements and photodetectors is optically interconnected with a vertical cavity surface-emitting laser array. We give examples of some algorithms the system can perform. Compact gradient index lenses are utilized and we compare their performance with conventional imaging systems.
As an alternative to the hypercube, the binary de Bruijn (BdB) network is recently receiving much attention. The BdB not only provides a logarithmic diameter, fault-tolerance, and simple routing but also requires fewe...
详细信息
As an alternative to the hypercube, the binary de Bruijn (BdB) network is recently receiving much attention. The BdB not only provides a logarithmic diameter, fault-tolerance, and simple routing but also requires fewer links than the hypercube for the same network size. Additionally, a major advantage of the BdB network is a constant node degree: the number of edges per node is independent of the network size. This makes it very desirable for large-scale parallel systems. However, due to its asymmetrical nature and global connectivity it is posing a major challenge for VLSI technology. Optics, owing to its three-dimensional and global connectivity nature seems to be very suitable for implementing BdB networks. In this paper, we present an implementation methodology for optical BdB networks. The distinctive feature of the proposed implementation methodology is partitionability of the network into a few primitive operations that can be efficiently implemented. We further show feasibility of the presented design methodology by proposing an optical implementation of the BdB network.
Principle and design of a self-routing 36 by 36 free-space optically interconnected optoelectronic crossbar cell-switch with a high bit rate and reconfigurable between two successive cells, is detailed. Derived from A...
详细信息
Principle and design of a self-routing 36 by 36 free-space optically interconnected optoelectronic crossbar cell-switch with a high bit rate and reconfigurable between two successive cells, is detailed. Derived from ATM oriented telecom switch and optical computing it uses extensively the vertical cavity surface emitting lasers and their polarization control both electrical and optical. An original fractal hexagonal geometry is proposed which enables use of low cost optically recorded holograms and compact integrated optical devices.
We propose a novel interconnection network usingoptical datalinking and switching technologies for shared memory multiprocessors. A splitter-combiner space division optical switch using semiconductor optical amplifie...
详细信息
We propose a novel interconnection network usingoptical datalinking and switching technologies for shared memory multiprocessors. A splitter-combiner space division optical switch using semiconductor optical amplifier gates is a key element. The experimental results show that the scale of a single-stage optical switch with throughput of 10 Gbit/s per a port is extended up to 16×16. The results of numerical simulations show that the latency is less than 100 ns for the 16×16 interconnection network. These characteristics are remarkable in that electrical switches have never achieved such large throughput, high scaleability and small latency simultaneously. An interconnection network usingoptical switching technologies is a novel platform for high performance data communications in shared memory multiprocessors.
This paper demonstrates that the real-time utilization of image sequences, at flame rates far above what is currently possible, can now be achieved with an optically interconnected massivelyparallel processor. A foca...
详细信息
This paper demonstrates that the real-time utilization of image sequences, at flame rates far above what is currently possible, can now be achieved with an optically interconnected massivelyparallel processor. A focal plane processing chip with an on-chip array of sigma-delta analog to digital converter front ends under each pixel is presented. This two layer chip is a scaleable high frame rate image capture building block, however it requires a third layer of data processing to filter the sigma-delta front end data to obtain images. The use of an array of optically connected processors beneath the chip proves to be the best solution to this challenging data processing task.
A new way to dynamically control in-flight pulses by a co-propagating shepherd pulse in a wavelength division multiplexed (WDM) single-mode fiber system was proposed at the MPPOI '96 conference. That system functi...
详细信息
ISBN:
(纸本)0818679751
A new way to dynamically control in-flight pulses by a co-propagating shepherd pulse in a wavelength division multiplexed (WDM) single-mode fiber system was proposed at the MPPOI '96 conference. That system functionally resembles an optical fiber ribbon cable, except that all the bits pass on one fiber optic waveguide. This single fiber bit parallel wavelength link can be used to extend the (speed x distance) product of emerging cluster computer networks, such as, the MyriNet, SCI, Hippi-6400, ShuffleNet, etc. Here, we shall present the first experimental evidence that this pulse shepherding effect can be observed in a commercially Corning DS (dispersion-shifted) fiber simulation results will first be presented for the case observed in the laboratory setup. A discussion of the experiment setup and measurement procedures will be given. Experimental results will then be compared with computer generated results. Excellent agreement is observed. Future experiments dealing with the shepherding effect among more than two co-propagating pulses will be performed.
As the performance of individual elements within parallelprocessing systems increases, increased communication capability between distributed processor and memory elements is required. There is great interest in usin...
详细信息
As the performance of individual elements within parallelprocessing systems increases, increased communication capability between distributed processor and memory elements is required. There is great interest in using fiber optics to improve interconnect communication beyond that attainable using electronic technology. Several groups have considered WDM, star-coupled optical interconnects. In this paper, we propose a fiber optic transceiver to provide low latency, high bandwidth channels for such interconnects using a robust multimode fiber technology. We use instruction-level simulation to quantify the bandwidth, latency, and concurrency required for such interconnects to scale to 256 nodes, each operating at 1 GFLOPS performance. We show that performance scales to 100 GFLOPS for scientific application kernels using a small number of wavelengths (8 to 32), only one wavelength received per node, and achievable optoelectronic bandwidth and latency.
In shuffle multistage interconnection networks (MINs), by increasing the size of the switches, the number of stages (NS) and in turn the attenuation of signals decreases. An interesting concept for self-routing in shu...
详细信息
ISBN:
(纸本)0818679751
In shuffle multistage interconnection networks (MINs), by increasing the size of the switches, the number of stages (NS) and in turn the attenuation of signals decreases. An interesting concept for self-routing in shuffle MINs are dimension-dependent switches, i.e. switches of size 2(d) x 2(d) (where d greater than or equal to 1 is the dimension of the interconnected data set) and dimension-dependent interconnections. Throughout the paper, the 1-D shuffle interconnections and k x k-switches (k greater than or equal to 4) are replaced by d-dimensional shuffles (d greater than or equal to 2) and switches of size 2(d) x 2(d) though the networks are still planar. In this way, given a data set, the original contribution of the paper is the presentation of (1) the number of topologically equivalent shuffle networks, (2) the complete concept of self-routing in (artificial) higer-dimensional shuffle networks (3) the solution of self-routing in 2-D shuffle networks not solvable by the common concept and (4) the decomposition of the input/output array into several small arrays without loss of functionality. The latter is important as presently reliable laser arrays have size of at most 8 x 8 and 16 x 16, respectively.
暂无评论