Stefanos Kaxiras, "Kiloprocessor Extensions to SCI." In Proceedings of the 10th International Parallel Processing Symposium, April 1996. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
High Performance Switch Architectures For CC-Numa Multiprocessors - Iyer (1999) (Correct)
....of the switch directory are measured by running several scientific applications on an execution driven simulator and commercial workloads on a trace driven simulator. Related work in this area includes the design of directory caches [51] the MIND network [57] and the GLOW STEM extensions to SCI [32, 33]. Michael et al. 51] recently presented design issues for directory caches to improve the performance of applications. They consider SRAM and DRAM directory designs and present the performance of several scientific applications in a CC NUMA ennvironment. The MIND network [57] embeds directories ....
....network [57] embeds directories within the switches of the MIN and presents a hierarchical directory protocol to maintain cache coherence. This scheme suffers from restrictions imposed by the inclusion property requiring larger directories towards the root of the hierarchy. Finally, Kaxiras et al. [32, 33] present two extensions to the SCI protocol called GLOW and STEM to provide scalable reads and writes. They employ directories in the switches to detect widely shared data. Our switch directory framework is aimed at improving the performance of communication intensive applications by reducing the ....
[Article contains additional citation context not shown here]
S. Kaxiras and J. Goodman, "Kiloprocessor Extensions to SCI," 10th International Parallel Processing Symposium, Apr. 1996. 149
Identification And Optimization Of Sharing Patterns For Scalable.. - Kaxiras (1998) (4 citations) Self-citation (Kaxiras) (Correct)
....forwarding The fully SCI compatible GLOW extensions use SCI s invalidation mechanism to invalidate their children and use request forwarding amongst them. All the GLOW agents in parallel assume a role similar to the head nodes in SCI and invalidate their child lists using the SCI invalidation [49]. The invalidation of a GLOW sharing tree is shown in Figure 3.16 (the initial tree is shown in Figure 3.16A) After the cache block is written, the node starts invalidating the highest level of the sharing tree using the SCI invalidation protocol (Figure 3.16B, transaction 1) However, the GLOW ....
Stefanos Kaxiras, "Kiloprocessor Extensions to SCI." In Proceedings of the 10th International Parallel Processing Symposium, April 1996.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC