A Task Migration System for Parallel Scientific Computations in Heterogeneous NOW Environments.
Chung-Shang Shao, Bobby Schnabel · 1999
This paper presents an source-level software system, PMT, which performs task migrations for long-running parallel scientific computations among a network of architecturally different workstations. As machine loads vary through time in a nondedicated environment, application performance can be improved by utilizing task migrations. This system is designed for parallel applications written in the C language, which run on PVM. It employs an independent checkpointing and restart mechanism for migration in heterogeneous environments. In addition, the system does not require any modification of the underlying computing environments for portability. Preliminary experiments show that the system introduces low overheads on normal executions when no migration occurs. Its migration performance is reasonably efficient for supporting long-running applications. 1 Introduction The network of workstations (NOW) environment has drawn a lot of attention in the research community. Various systems [1] [2] ...