|Home \ Graduation Activities \ Post-Graduation Page||Login|
Clustering e Scheduling Distribuído de VMs
In recent years, computer clusters made entirely of simple desktop computers are becoming the standard for high-performance computing, as the scalability and cost-effiency of such solution surpasses most high-end-mainframes. If the workstations in a cluster can work collectively and provide the illusion of being a single workstation with more resources, then we would have what is referred in the literature as a Single System Image. Much research has been done in this area, such as Distributed Shared Memory systems and Distributed Virtual Machines that can run applications written in a high-level language in a cluster, behaving as if it were on a single machine. In this work, we present Caft, a middleware that runs on top of the Terracotta system and has the capacity to run simple multi-threaded Java applications in a transparent way, scheduling threads across the several nodes in a Terracotta cluster and taking advantage of the computational and memory resources available in the cluster. We use bytecode instrumentations to add clustering capabilities to the multi-threaded Java application, as well as extra synchronization if needed. The middleware supports several modes, in order achieve a balance between transparency and flexibility.
Start Date: 2010-11-04
End Date: 2010-11-04
Post-Graduation Student / Researcher / Professor: