Performance Impact of Code and Data Placement on the IBM RP3 (1989) [1 citations — 0 self]
Popular Tags
No tags have been applied to this document.
BibTeX | Add To MetaCart
author = {Mats Brorsson},
title = {Performance Impact of Code and Data Placement on the IBM RP3},
year = {1989}
}
Bookmarks
OpenURL
Abstract:
The IBM RP3 belongs to a class of shared-memory multiprocessors with non-uniform memory access times. Some memory locations are more costly to access than others. Most commercial multiprocessors have uniform memory access times. Placement of instructions, private data and shared data can significantly alter program performance on non-uniform access machines. In this paper we discuss a series of measurements comparing the performance impact of alternative code and data placement strategies on a 64-node RP3. Page replication is used as a method of placing shared memory in local storage. Virtual pages are replicated to local memory on demand, ensuring that only those pages actually used will occupy local storage. The results from the RP3 experiments confirm earlier reported results but extend these results to a higher degree of parallelism. Localizing code and private data is crucial to obtain reasonable speedup in the absence of cache memories. With processor caches, code can reside in g...
Citations
| 53 | A single-program-multipledata computational model for EPEX/FORTRAN – Darema, George, et al. - 1988 |
| 25 | RP3 Processor-Memory Element – Branticy, McAuliffe, et al. - 1985 |
| 6 | Reducing Contention in Shared-Memory Multiprocessors – Stenstrom - 1988 |
| 4 | The RP3 Parallel Computing Environment – Bryant - 1988 |
| 4 | Analysis of Cache Memories in Highly Parallel Systems – McAuliffe - 1986 |
| 4 | Performance evaluation of multiple processor systems – Raskin - 1978 |
| 2 | A Study of the Memory Reference Behaviour of Engineering /Scientific – Baylor, Rathi - 1989 |
| 1 | Accetta et al. Mach: A New Kernel Foundation for Unix Development – R - 1986 |
