Department of Computer Science
Researcher:
Dr. Cho-Li Wang, Associate Professor (clwang@cs.hku.hk)
Project Title:
G-JavaMPI: A Grid Middleware for Lightweight MPI Task Migration
Project Description & Significance:
G-JavaMPI is an implementation of Java binding of the MPI standard for computation grids. It enables efficient message passing between distributed Java processes by exploiting the MPI technologies. In addition, it supports transparent process migration. With the use of JVMTI and bytecode instrumentation, it does not require any modification of the JVM, and hence provides better portability across different JVM implementations. Though portable task migration support allows better resource utilization and data access locality in grid computing, existing task migration mechanisms are however too coarse-grained and costly, offsetting the benefits from migration and hampering flexible task partitioning among grid resources. In order to further leverage the power of grid computing, we introduce a new computation migration technique called stack-on-demand (SOD) that exports partial execution states of a stack machine to achieve agile mobility, easing into small-capacity devices and flexible distributed execution in a multi-domain workflow style. SOD paves the way for grid computing with the features of lightweight task migration, distributed workflow style and autonomous task roaming. We have integrated SOD into G-JavaMPI middleware. It is shown that SOD migration cost is pretty low, comparing to several existing migration mechanisms. We also conduct experiments with an iPhone handset to demonstrate the elasticity of SOD by which heavyweight server-side processes can run adaptively on the cell phone.
 |
 |
