Combining OAR with the power of volunteer computing through BOINC

Description:

In the current scenario, we can observe two distinct class of resources that are used for grid computing. Each of these classes has its own advantages and advantages. On one hand, there are high-end, clusters or microprocessors that provide rich-feature sets dedicated to computation of user submitted jobs, whereas on the other hand, we have Desktop grids that combine the power of thousands of light-weight systems (desktops) and utilize their idle times to execute applications. However, in order to fully harness the potential of grid computing, it is necessary to develop an interface that can communicate with both these classes with equal ease, in order to use the best features offered by each ofthese platform types. Hence, in this project, we try to integrate two different resource managementsystems. These are namely the Berkeley Open Infrastructure for Network Computing (BOINC), which is a resource management middleware for volunteer computing and desktop grid computing, and OAR, which a resource manager (batch scheduler) for large clusters and Grids. We implement a B-OAR interface that understands the underlying architectures of both these systems and translates the data storage and interpretation formats, in order to make the two systems communicate.

Status:

Jobs submitted through OAR can run on volunteer resources through a BOINC server.

Future work:

We are currently looking for real compute-intensive applications to utilize this new tool. We are also implementing virtual resources in OAR to give a user the perception of a dedicated platform built from volatile resources.

Developers:

Nagarjun Kota, Olivier Richard, Derrick Kondo with much guidance from the OAR team.

Contact:

Derrick Kondo (derrick.kondo :: inria.fr)