Title:
Min-cut Methods for Mapping Dataflow Graphs
Min-cut Methods for Mapping Dataflow Graphs
Authors
Elling, Volker Wilhelm
Schwan, Karsten
Schwan, Karsten
Authors
Person
Advisors
Advisors
Associated Organizations
Organizational Unit
Series
Collections
Supplementary to
Permanent Link
Abstract
High performance applications and the underlying hardware platforms
are becoming increasingly dynamic; runtime changes in the
behavior of both are likely to result in inappropriate mappings of
tasks to parallel machines during application execution. This fact is
prompting new research on mapping and scheduling the dataflow graphs that
represent parallel applications. In contrast to recent research which
focuses on critical paths in dataflow graphs, this paper presents new
mapping methods that compute near-min-cut partitions of the dataflow graph.
Our methods deliver mappings that are an order of magnitude more
efficient than those of DSC, a state-of-the-art critical-path algorithm,
for sample high performance applications.
Sponsor
Date Issued
1999
Extent
247669 bytes
Resource Type
Text
Resource Subtype
Technical Report