Hierarchical Deployment for the GRID

Distributed computing is moving from clusters to Grids and P2P. A cluster is rather easy to use due to its homogeneous nature and latitudinarian internal network policy. Grids and P2P networks involve multiple heterogeneous sites, probably not fully interconnected, and subject to various network or security policies. One big challenge with these environments is to be able to deploy and allow, efficient, communication between all workers. This Research Master proposes to tackle those difficult issues.

A typical workaround is to use a master/worker model. The master can reach all workers and each worker can reach the master. Workers do not, or cannot, talk between them. This solution is indeed very limiting, preventing many applications to be deployed and executed. Hierarchical deployment and message routing are an attempt to solve this issue by providing to users a transparent mechanism to allow bidirectional communications between each worker, whatever encountered network restrictions (firewall, NAT, private IPs etc.).

The research takes place in Sophia Antipolis, within the OASIS research team, a joint project between INRIA, CNRS-I3S, and University of Nice Sophia Antipolis.

The OASIS project conducts research on distributed objects. The team has been designing and implementing  ProActive, a Java library for the GRID. In the framework of an active object model, the library features Asynchronous Typed Messages, Future Synchronizations (Wait-by-necessity), Group Communications, Mobility, Security, and a generic component model. At the infrastructure level, XML Deployment Descriptors provide the capacity to deploy on many kind of GRIDs.
In the absence of any syntactical extension, ProActive programmers write standard code. The library is itself extensible by the programmers, using a MOP (Meta-Object Protocol), making the system open for adaptations and optimizations.

Overall, the following technical features are expected:

• efficient, dynamic, adaptive communications (including non tree-based routing),
• fault-tolerance upon the failure of some forwarder nodes,
• mixing of static and dynamic strategies.
  

Overall, the work will include conceptual models and practical experiments. The following steps could be followed:

• Analysis of existing models, techniques, and tools for hierarchical deployment and message routing,
• Analysis of generic tools for Grid computing, active object, component models, ProActive infrastructure
• Proposition of an architecture for full fletched  hierarchical deployment and message routing
• Definition of a framework to achieve this model within the ProActive platform
• Programming of an efficient implementation, experiments and benchmarks

Nice Sophia Antipolis Advisor :  Denis Caromel
Téléphone : 04 92 38 76 31 Email : Denis.Caromel@inria.fr
Laboratoire ou équipe : INRIA Sophia Antipolis -- I3S -- CNRS

Hardware and software to be used : Networks of PCs, Clusters, Intranet and Internet P2P machines

Internship location:
       Sophia Antipolis, between Nice and Cannes, France