Fault tolerance based on the publish-subscribe paradigm for the BonjourGrid middleware

Abstract

How to federate the machines of all Boinc, Condor and XtremWeb projects? If you believe in volunteer computing and want to share more than one project then BonjourGrid may help. In previous works, we proposed a novel approach, called BonjourGrid, to orchestrate multiple instances of Institutional Desktop Grid middleware. It is our way to remove the risk of bottleneck and failure, and to guarantee the continuity of services in a distributed manner. Indeed, BonjourGrid can create a specific environment for each user based on a given computing system of his choice such as XtremWeb, Condor or Boinc. This work investigates, first, the procedure to deploy Boinc and Condor on top of BonjourGrid and, second, proposes a fault tolerant approach based on passive replication and virtualization to tolerate the crash of coordinators. The novelty resides here in an integrated environment based on Bonjour (publication-subscription mecanism) for both the coordination protocol and for the fault tolerance issues. In particular, it is not so frequent to our knowledge to describe and to implement a fault tolerant protocol according to the pub-sub paradigm. Experiments, conducted on the Grid'5000 testbed, illustrate a comparative study between Boinc (respectively Condor) on top of BonjourGrid and a centralized system using Boinc (respectively Condor) and second prove the robustness of the fault tolerant mechanism.