Communication and computation performance of the CM-5

Kwan, T.T.; Totty, Brian; Reed, Daniel A.

doi:10.1145/169627.169691

Cited by 25 publications

(8 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The same fast T3D communications are not found on the CM-5 system: transfer times and latency overheads are measured to be ten and 30 times worse than the respective shared memory T3D ones. The presented results are in agreement with [26] for the CM-5 (a worse bandwidth is found in [23], probably due to the old CMMD version), with [25] for the T3D, and with [20,22] for the SP2 (complete results for several MPP systems can be found in [25]). Communication performance, indeed, can be considered the true bottleneck for CM-5 applications (the improvement achieved by low-level message passing libraries on the CM-5 is shown in [23]).…”

Section: Latency and Bandwidthsupporting

confidence: 86%

“…The maximum bidirectional bandwidths measured in the shift operation on the three MPP systems are reported in Table 2. As in the previous experiment, the T3D within the shared memory model achieves very high transfer rates, outperforming by a factor of three and ten, respectively, the SP2 with MPL (in agreement with [20]) and the CM-5 (in agreement with [26]). We want to stress that the best performance obtained on the T3D is twice the W measured in the ping-pong operation (almost 50 and 110 Mbyte/s of peak are reached with PVMfast and shared memory primitives).…”

Section: Global Shift Operationsupporting

confidence: 73%

See 1 more Smart Citation

Benchmark kernels as a tool for performance evaluation of MPP‘s

Marenzoni

Rossi²

1998

Concurrency: Pract. Exper.

View full text Add to dashboard Cite

Section: Latency and Bandwidthsupporting

confidence: 86%

Section: Global Shift Operationsupporting

confidence: 73%

Benchmark kernels as a tool for performance evaluation of MPP‘s

Marenzoni

Rossi²

1998

Concurrency: Pract. Exper.

View full text Add to dashboard Cite

“…This is because as the number of processors increases, the dataparallel operations supported by pipeline bits across the long wires are much less distance sensitive. The experiments in [9] show that message transmission latencies and bandwidths are independent of the partition size on the CM-5. The network latencies vary only slightly with the number of network levels crossed.…”

Section: Network Latency Patterns Versus Cm-5 Fat-tree System Scalingmentioning

confidence: 93%

Comparative Evaluation and Case Studies of Shared‐Memory and Data‐Parallel Execution Patterns

Zhang

Sun²

1999

Scientific Programming

View full text Add to dashboard Cite

USAShared-memory and data-parallel programming models are two important paradigms for scientific applications. Both models provide high-level program abstractions, and simple and uniform views of network structures. The common features of the two models significantly simplify program coding and debugging for scientific applications. However, the underlining execution and overhead patterns are significantly different between the two models due to their programming constraints, and due to different and complex structures of interconnection networks and systems which support the two models. We performed this experimental study to present implications and comparisons of execution patterns on two commercial architectures. We implemented a standard electromagnetic simulation program (EM) and a linear system solver using the shared-memory model on the KSR-1 and the data-parallel model on the CM-5. Our objectives are to examine the execution pattern changes required for an implementation transformation between the two models; to study memory access patterns; to address scalability issues; and to investigate relative costs and advantages/disadvantages of using the two models for scientific computations. Our results indicate that the EM program tends to become computationintensive in the KSR-1 shared-memory system, and memorydemanding in the CM-5 data-parallel system when the systems and the problems are scaled. The EM program, a highly data-parallel program performed extremely well, and the linear system solver, a highly control-structured program suffered significantly in the data-parallel model on the CM-5. Our study provides further evidence that matching execution patterns of algorithms to parallel architectures would achieve better performance.

show abstract

“…Most of the literature deals with the CM-5 and focuses on raw network performance. 22 ' 27 ' 29 Typical communication patterns include simple sends and pingpong between pairs of nodes. Block permutations of data and grid shifts have been shown to have little or no contention on the CM-5.…”

Section: Introductionmentioning

confidence: 99%

PERFORMANCE ANALYSIS OF WORMHOLE ROUTED K-Ary N-TREES

Petrini

Vanneschi

1998

Int. J. Found. Comput. Sci.

View full text Add to dashboard Cite

The past few years have seen a rise in popularity of massively parallel architectures that use fat-trees as their interconnection networks. In this paper we formalize a parametric family of fat-trees, the fc-ary n-trees, built with constant arity switches interconnected in a regular topology. A simple adaptive routing algorithm for fc-ary n-trees sends each message to one of the nearest common ancestors of both source and destination, choosing the less loaded physical channels, and then reaches the destination following the unique available path. Through simulation on a 4-ary 4-tree with 256 nodes, we analyze some variants of the adaptive algorithm that utilize wormhole routing with 1, 2 and 4 virtual channels. The experimental results show that the uniform, bit reversal and transpose traffic patterns are very sensitive to the flow control strategy. In, all these cases, the saturation points are between 35 -40% of the network capacity with 1 virtual channel, 55-60% with 2 virtual channels and around 75% with 4 virtual channels. The complement traffic, a representative of the class of the congestion-free communication patterns, reaches an optimal performance, with a saturation point at 97% of the capacity for all flow control strategies. In this case virtual channels are of little help and the average network latency experiences an increase proportional to the number of virtual channels, due to the multiplexing of several packets onto the same physical links.

show abstract

Communication and computation performance of the CM-5

Abstract: The ThinkingMachines CM-5

Cited by 25 publications

References 8 publications

Benchmark kernels as a tool for performance evaluation of MPP‘s

Benchmark kernels as a tool for performance evaluation of MPP‘s

Comparative Evaluation and Case Studies of Shared‐Memory and Data‐Parallel Execution Patterns

PERFORMANCE ANALYSIS OF WORMHOLE ROUTED K-Ary N-TREES

Contact Info

Product

Resources

About