Download:
|
by David R. Surma, Edwin H-m. Sha, Nelson Passos
ftp://wizard.cse.nd.edu/pub/Reports/1997/tr-97-14.ps.gz
Add To MetaCart
Abstract:
In massively parallel systems, the performance gains are often significantly diminished by the inherent communication overhead. This overhead is caused by the required message passing resulting from the task allocation scheme. In this paper, techniques to reduce this communication overhead by both scheduling the communication and determining the routing that the messages should take within a tightly-coupled processor network are presented. Using the recently developed Collision Graph model, static scheduling algorithms are derived which work at compiletime to determine the ordering and routing of the individual message transmissions. Since a priori knowledge about the network traffic required by static scheduling may not be available or accurate, this work also considers dynamic scheduling. A novel hybrid technique is presented which operates in a dynamic environment yet uses known information obtained by analyzing the communication patterns. Experiments performed show significant improvement over baseline techniques. Index Terms- Tightly-coupled networks, communication scheduling, parallel systems, graph modeling, routing.
Citations
|
7709
|
Computers and Intractability: A Guide to the Theory of NP-Completeness
– Garey, Johnson
- 1979
|
|
5824
|
Introduction to Algorithms
– Cormen, Leiserson, et al.
- 1990
|
|
793
|
Computer Networks
– Tanenbaum
- 1996
|
|
423
|
A Survey of Wormhole Routing Techniques in Direct Networks
– Ni, McKinley
- 1993
|
|
224
|
Introduction to Sequencing and Scheduling
– Baker
- 1974
|
|
145
|
Theory of Scheduling
– Conway, Maxwell, et al.
- 1967
|
|
93
|
Collective Communication in Wormhole-Routed Massively Parallel
– McKinley, Tsai, et al.
- 1995
|
|
81
|
Partitioning and mapping algorithms into fixed size systolic arrays
– Moldovan, Fortes
- 1986
|
|
75
|
Practical multiprocessor scheduling algorithms for efficient parallel processing
– Kasahara, Narita
- 1984
|
|
46
|
EXECUBE - a new architecture for scalable MPPs
– Kogge
- 1994
|
|
39
|
Priority based real-time communication for large scale wormhole networks
– Li, Mutka
- 1994
|
|
28
|
Interprocessor Traffic Scheduling Algorithm for Multiple-Processor Networks
– Bianchini, Shen
- 1987
|
|
23
|
A compile-time technique for controlling real-time execution of task-level data-flow graphs
– Shukla, Little
- 1992
|
|
17
|
Path selection for communicating tasks in a wormhole-routed multicomputer
– Lee, Kim
- 1994
|
|
17
|
Traffic routing for multicomputer networks with virtual cut-through capability
– Kandlur, Shin
- 1992
|
|
14
|
Scheduling sequential
– Munshi, Simons
- 1990
|
|
12
|
Deadlock-free message routing inmultiprocessor interconnection networks
– Dally, Seitz
- 1987
|
|
9
|
Contention-free communication scheduling on 2d meshes
– Eberhart, Li
- 1996
|
|
9
|
Application specific communication scheduling on parallel systems
– Surma, Sha
- 1995
|
|
8
|
Static communication scheduling for minimizing collisions in application-specific parallel systems
– Surma, Sha
- 1996
|
|
8
|
Collision graph based communication scheduling for parallel systems
– Surma, Sha
- 1997
|
|
7
|
Optimal communication scheduling based on collision graph model
– Surma, Tongsima, et al.
- 1996
|
|
6
|
Efficient communication scheduling with re-routing based on collision graphs
– Surma, Sha
- 1997
|
|
6
|
SCORE: An Efficient Technique to Reduce Congestion in Parallel Systems
– Surma, Sha
- 1997
|
|
3
|
Hybrid static-dynamic communication scheduling for parallel systems
– Surma, Sha
- 1997
|
