Z. Lahjoumri and T. Priol. KOAN: A Shared-Memory for the iPSC/2 Hypercube. In CONPAR /VAPP92, LNCS 634. Springer-Verlag, September 1992. Home/Search Document Not in Database Summary Related Articles Check |
This paper is cited in the following contexts:
Page-level Affinity Scheduling for Eliminating False Sharing - Bodin, Granston, Montaut (Correct)
....sharing. False sharing is particularly problematic in software DSMs where coherency is maintained at the granularity of a page. Because software DSMs are typically implemented on top of or embedded into systems which were tuned for uniprocessor environments, page sizes are typically 4K or larger [LP92, KDCZ94] In this paper, we present a technique for eliminating false sharing by applying page level affinity scheduling. This technique is a generalization of our earlier work [Gra93, BGM94] which handled certain cases of one and two dimensional loops. The remainder of this paper is organized as ....
....is enforced without synchronization. 6 Experimental Results Page level affinity scheduling (PLAS) except for the precomputation, has been implemented in the FortranS compiler [BKP93] The Fortran S compiler generates SPMD style code that runs on the iPSC 2 under the KOAN software DSM [LP92] The KOAN DSM system is embedded in the operating system of the iPSC 2. Pages of size 4 KB are physically distributed across processors local memories. KOAN uses a distributed manager algorithm based on [Li86] with an invalidation protocol that ensures that the shared memory is coherent at all ....
[Article contains additional citation context not shown here]
Z. Lahjoumri and T. Priol. KOAN: A Shared-Memory for the iPSC/2 Hypercube. In CONPAR /VAPP92, LNCS 634. Springer-Verlag, September 1992.
Loop Transformations to Prevent False Sharing - Bodin, Granston, Montaut (1995) (4 citations) (Correct)
.... of Loop Nest 1 as a function of the time between successive writes by a given processor: ffl the original loop (ORIG) ffl a version optimized to prevent multiple writer false sharing (FS OPT) Both versions were executed on 16 processors of a 32 processor iPSC 2 under the KOAN SVM system [LP92] which supports the aforementioned invalidation based coherence protocol and employs a page size of 4 KB (512 double precision numbers) The problem size was N 1 = N 2 = 10 3 . The performance difference between the ORIG and FS OPT versions demonstrates the performance degradation that can ....
.... approach with our ping pong reduction transformation described in [BGM94] This hybrid strategy has been shown to work well in practice [BGM94] 8 Experimental Results FS OPT has been implemented in the Fortran S compiler [BKP93] which generates code that runs on the iPSC 2 under the KOAN SVM [LP92] The KOAN SVM system is embedded in the operating system of the iPSC 2. Pages of size 4 KB are physically distributed across processors local memories. KOAN uses a distributed manager algorithm based on [Li86] with an invalidation based protocol that ensures that the shared memory is coherent ....
[Article contains additional citation context not shown here]
Z. Lajormi and T. Priol. KOAN: A Shared-Memory for the iPSC/2 Hypercube. In CONPAR /VAPP92, LNCS 634. Springer-Verlag, September 1992.
Combining Flow and Dependence Analyses to Expose Redundant .. - Granston, Veidenbaum (1995) (1 citation) (Correct)
....analysis, flow analysis of parallel programs, dependence analysis, and data reuse. 1 Introduction Large scale, high performance, hierarchical and distributed shared memory systems are currently gaining in popularity. Although research prototypes of such systems have existed for some time [GKLS83, LP92, LLJ 92, KDCZ94] recently commercial versions such as the Kendall Square Research KSR1 and KSR2 [Ken92] the Cray T3D [Cra93] and the CONVEX Exemplar [CON93] have appeared on the market, and plans to build others have been announced. Although there is significant architectural variation ....
Z. Lajormi and T. Priol. KOAN: A Shared-Memory for the iPSC/2 Hypercube. In CONPAR /VAPP92, LNCS 634. Springer-Verlag, September 1992.
Improving the Performance of DSM Systems via Compiler Involvement - Mirchandaney (1994) (24 citations) (Correct)
....than one processor. The actual packing of structures to fit in a page (or a set of pages) so as to reduce false sharing was done manually through modifications to the program s data structures. No compiler support was available for iteration space tiling or array layouts. Projects such as KOAN [28] and others [9] perform optimizations to reduce the effect of false sharing on software DSM systems. One of the problems that some DSM systems face is the ping pong problem, where a page that is concurrently written by multiple writers moves rapidly between the processors. KOAN tries to minimize ....
Z. Lajormi and E. Priol. Koan: Shared memory for the ipsc/2 hypercube. In CONPAR-VAPP 92. SpringlerVerlig, September 1992.
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