Warp: an integrated solution of high-speed parallel computing
Proceedings of the 1988 ACM/IEEE conference on Supercomputing
Computer
Exploiting task and data parallelism on a multicomputer
PPOPP '93 Proceedings of the fourth ACM SIGPLAN symposium on Principles and practice of parallel programming
Experiments with a gigabit neuroscience application on the CM-2
Proceedings of the 1993 ACM/IEEE conference on Supercomputing
Architecture and evaluation of a high-speed networking subsystem for distributed-memory systems
ISCA '94 Proceedings of the 21st annual international symposium on Computer architecture
Software support for outboard buffering and checksumming
SIGCOMM '95 Proceedings of the conference on Applications, technologies, architectures, and protocols for computer communication
Gigabit I/O for distributed-memory machines: architecture and applications
Supercomputing '95 Proceedings of the 1995 ACM/IEEE conference on Supercomputing
A Host Interface Architecture for High-Speed Networks
Proceedings of the IFIP TC6/WG6.4 Fourth International Conference on High Performance Networking IV
Gigabit I/O for distributed-memory machines: architecture and applications
Supercomputing '95 Proceedings of the 1995 ACM/IEEE conference on Supercomputing
Network-Based Multicomputers: A Practical Supercomputer Architecture
IEEE Transactions on Parallel and Distributed Systems
A high-speed network interface for distributed-memory systems: architecture and applications
ACM Transactions on Computer Systems (TOCS)
| Hi-index | 0.00 |
We evaluate the impact of a gigabit network on the implementation of a distributed chemical process optimization application. The optimization problem is formulated as a stochastic Linear Assignment Problem and was solved using the Thinking Machines CM-2 (SIMD) and the Cray C-90 (vector) computers at PSC, and the Intel iWarp (MIMD) system at CMU, connected by the Gigabit Nectar testbed. We report our experience distributing the application across this heterogeneous set of systems and present measurements that show how the communication requirements of the application depend on the structure of the application. We use detailed traces to build an application performance model that can be used to estimate the elapsed time of the application for different computer system and network combinations. Our results show that the application benefits from the high-speed network, and that the need for high network throughput is increasing as computer systems get faster. We also observed that supporting high burst rates is critical, although structuring the application so that communication is overlapped with computation relaxes the bandwidth requirements.