Hyperplane grouping and pipelined schedules: How to execute tiled loops fast on clusters of SMPs

Athanasaki, M; Sotiropoulos, A; Tsoukalas, G; Koziris, N; Tsanakas, P

dc.contributor.author	Athanasaki, M	en
dc.contributor.author	Sotiropoulos, A	en
dc.contributor.author	Tsoukalas, G	en
dc.contributor.author	Koziris, N	en
dc.contributor.author	Tsanakas, P	en
dc.date.accessioned	2014-03-01T01:22:29Z
dc.date.available	2014-03-01T01:22:29Z
dc.date.issued	2005	en
dc.identifier.issn	0920-8542	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/16586
dc.subject	Hyperplanes	en
dc.subject	Loop tiling	en
dc.subject	Pipelined schedules	en
dc.subject	Supernodes	en
dc.subject	Tile grouping	en
dc.subject.classification	Computer Science, Hardware & Architecture	en
dc.subject.classification	Computer Science, Theory & Methods	en
dc.subject.classification	Engineering, Electrical & Electronic	en
dc.subject.other	Algorithms	en
dc.subject.other	Bandwidth	en
dc.subject.other	Communication systems	en
dc.subject.other	Computer networks	en
dc.subject.other	Data acquisition	en
dc.subject.other	Interfaces (computer)	en
dc.subject.other	Storage allocation (computer)	en
dc.subject.other	Virtual reality	en
dc.subject.other	Hyperplanes	en
dc.subject.other	Loop tiling	en
dc.subject.other	Pipelined schedules	en
dc.subject.other	Supernodes	en
dc.subject.other	Tile grouping	en
dc.subject.other	Parallel processing systems	en
dc.title	Hyperplane grouping and pipelined schedules: How to execute tiled loops fast on clusters of SMPs	en
heal.type	journalArticle	en
heal.identifier.primary	10.1007/s11227-005-0298-8	en
heal.identifier.secondary	http://dx.doi.org/10.1007/s11227-005-0298-8	en
heal.language	English	en
heal.publicationDate	2005	en
heal.abstract	This paper proposes a novel approach for the parallel execution of tiled Iteration Spaces onto a cluster of SMP PC nodes. Each SMP node has multiple CPUs and a single memory mapped PCI-SCI Network Interface Card. We apply a hyperplane-based grouping transformation to the tiled space, so as to group together independent neighboring tiles and assign them to the same SMP node. In this way, intranode (intragroup) communication is annihilated. Groups are atomically executed inside each node. Nodes exchange data between successive group computations. We schedule groups much more efficiently by exploiting the inherent overlapping between communication and computation phases among successive atomic group executions. The applied non-blocking schedule resembles a pipelined datapath, where group computation phases are overlapped with communication ones, instead of being interleaved with them. Our experimental results illustrate that the proposed method outperforms previous approaches involving blocking communication or conventional grouping schemes. © 2005 Springer Science + Business Media, Inc.	en
heal.publisher	SPRINGER	en
heal.journalName	Journal of Supercomputing	en
dc.identifier.doi	10.1007/s11227-005-0298-8	en
dc.identifier.isi	ISI:000232722900003	en
dc.identifier.volume	33	en
dc.identifier.issue	3	en
dc.identifier.spage	197	en
dc.identifier.epage	226	en