Scalability of parallel machines

Nussbaum, Daniel A.; Agarwal, Anant

doi:10.1145/102868.102871

Cited by 93 publications

(32 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…It is proving somewhat difficult for standard (flat) interconnection network topologies to scale to large numbers of processors while satisfying these cost and performance constraints, so in recent years, there has been a strong interest in clustered hierarchical networks [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15]. Hierarchical networks take a modular approach to network design, where local clusters of processors are connected using an intracluster network.…”

Section: Scalable Optical Crossbar-connected Interconnection Networkmentioning

confidence: 99%

“…To this end, we propose a class of all-optical, highly scalable, hierarchical [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15] network architectures, referred to as the Scalable Optical Crossbar-Connected Interconnection Networks (SOCNs). A SOCN class network is a two-level hierarchical network that is based on a fully connected wavelength-division multiplexed (WDM) crossbar interconnect.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A class of highly scalable optical crossbar-connected interconnection networks (SOCNs) for parallel computing systems

Webb

Louri

2000

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

AbstractÐA class of highly scalable interconnect topologies called the Scalable Optical Crossbar-Connected Interconnection Networks (SOCNs) is proposed. This proposed class of networks combines the use of tunable Vertical Cavity Surface Emitting Lasers (VCSEL's), Wavelength Division Multiplexing (WDM) and a scalable, hierarchical network architecture to implement large-scale optical crossbar based networks. A free-space and optical waveguide-based crossbar interconnect utilizing tunable VCSEL arrays is proposed for interconnecting processor elements within a local cluster. A similar WDM optical crossbar using optical fibers is proposed for implementing intercluster crossbar links. The combination of the two technologies produces large-scale optical fan-out switches that could be used to implement relatively low cost, large scale, high bandwidth, low latency, fully connected crossbar clusters supporting up to hundreds of processors. An extension of the crossbar network architecture is also proposed that implements a hybrid network architecture that is much more scalable. This could be used to connect thousands of processors in a multiprocessor configuration while maintaining a low latency and high bandwidth. Such an architecture could be very suitable for constructing relatively inexpensive, highly scalable, high bandwidth, and fault-tolerant interconnects for large-scale, massively parallel computer systems. This paper presents a thorough analysis of two example topologies, including a comparison of the two topologies to other popular networks. In addition, an overview of a proposed optical implementation and power budget is presented, along with analysis of proposed media access control protocols and corresponding optical implementation.

show abstract

Section: Scalable Optical Crossbar-connected Interconnection Networkmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A class of highly scalable optical crossbar-connected interconnection networks (SOCNs) for parallel computing systems

Webb

Louri

2000

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

show abstract

“…In massively parallel systems parallelism can be exploited by performing concurrent operations. Though CREW access is a desirable property in parallel machines, the performance improvement can be easily attained by mapping the code to an underlying EREW machines [1]. There are several well-known, different fast parallel algorithms for matrix-matrix multiplication, but most of them use the CRCW (Concurrent Read Concurrent Write) model.…”

Section: Introductionmentioning

confidence: 99%

A New Parallel Algorithm for EREW PRAM Matrix Multiplication

Vollala

Geetha

Joshi

et al. 2017

Proceedings of the International Conference on Communication and Signal Processing 2016 (ICCASP 2016)

View full text Add to dashboard Cite

Abstract. This work presents a new parallel matrix multiplication algorithm using an exclusive-read and exclusive-write (EREW), parallel random access memory (PRAM) model for a fixed number of processors. This algorithm used for computing the matrix multiplication analyses the logical pattern that exists for accessing the elements of the matrix efficiently. The proposed algorithm PEMM (Parallel EREW algorithm for Matrix Multiplication) works with the time complexity of O(n), but takes less number of iterations when compared with the existing work. It can also run on parallel machines other than the EREW PRAM.

show abstract

“…A formal definition due to [22] focuses on algorithm-architecture scalability. The ratio of speedups of the algorithm on a real machine and a specific theoretical parallel machine (PRAM) defines scalability.…”

Section: Introductionmentioning

confidence: 99%

Architecture scalability of parallel vector computers with a shared memory

Dekker

1998

IEEE Trans. Comput.

View full text Add to dashboard Cite

Abstract-Based on a model of a parallel vector computer with a shared memory, its scalability properties are derived. The processor-memory interconnection network is assumed to be composed of crossbar switches of size b b. This paper analyzes sustainable peak performance under optimal conditions, i.e., no memory bank conflicts, sufficient processor-memory bank pathways, and no interconnection network conflicts. It will be shown that, with fully vectorizable algorithms and no communication overhead, the sustainable peak performance does not scale up linearly with the number of processors p. If the interconnection network is unbuffered, the number of memory banks must increase at least with O(p log b p) to sustain peak performance. If the network is buffered, this bottleneck can be alleviated; however, the half performance vector length still increases with O(log b p). The paper confirms the validity of the model by examining the performance behavior of the LINPACK benchmark.

show abstract

Scalability of parallel machines

Cited by 93 publications

References 4 publications

A class of highly scalable optical crossbar-connected interconnection networks (SOCNs) for parallel computing systems

A class of highly scalable optical crossbar-connected interconnection networks (SOCNs) for parallel computing systems

A New Parallel Algorithm for EREW PRAM Matrix Multiplication

Architecture scalability of parallel vector computers with a shared memory

Contact Info

Product

Resources

About