|
|
G-JavaMPI Project
Core members: Chen Lin, Ma Tianchi Supervisors: Dr. Cho-Li Wang and Dr. Francis Lau Project Description: The computational grid is analogous to electric power grid. It allows to couple geographically distributed resources and offers consistent and inexpensive access to resources irrespective of their physical location or access point. The principal difference between grid computing and the other paradigms is that grid resources are typically not exclusively dedicated to the grid usage. These resources may reside in different domains, have different configurations, and be connected by networks with fluctuating bandwidth or even disconnected temporarily. All these characteristics demand the development of a special Grid middleware to support efficient Grid computing. In this research, a new Grid middleware, named G-JavaMPI will be presented. This middleware extends the parallel computing capability of Java on the Grid with the support of a Grid-enabled message passing interface (MPICH-G2) for efficient communication between distributed Java processes. A special feature of G-JavaMPI is the support of transparent Java process migration for achieving global load balancing. In G-JavaMPI, the Java state capturing and restoring will be done through the standard Java Virtual Machine debugger interface (JVMDI). This makes it highly portable as it requires no modification of underlying OS or JVM. For achieving load balancing, we propose several communication-efficient process redistributing algorithms using the provided Java migration feature. All these algorithms consider communication costs based on the available network bandwidth between Grid points at run time. In particular, the decision of Java process migration will be made based on the prediction of future communication patterns of the target applications using bytecode parsing and execution tracing. The proposed G-JavaMPI middleware can help scientific or engineering applications adapt well to the dynamically changing network bandwidth and available computing resources in Grid, thereby, achieve higher performance. Related Work: Publication:
|