Section: New Results

Software

Fault-tolerance in KAAPI

We have developed a new algorithm to have a high performance fault tolerant mechanism in KAAPI. The protocol is based on coordinated checkpointing. The algorithm is well suited for iterative parallel application. The originality of our protocol is to allows partial restart of processes after detection of a fault.

Scalability of KAAPI

KAAPI software has been tested on two grid platforms, the French National Grid50000 and the Japanese Intrigger, during the 5th PLUGTEST event organised by ETSI and project OASIS at Sophia-Antipolis, France, Octobre, 20th - October, 24th, 2008. The KAAPI team took part of the Super Quant Monte Carlo contest during the PLUGTEST event and was the winner in front of 7 teams. Runs have shown the ability to fully exploit machines geographically distributed among France and Japan which demonstrates concrete communication between processes behind firewalls.

GRID5000: scheduling algorithm for OAR and authentication

OAR is a batch scheduler developed by Mescal team. The MOAIS team develops the central automata and the scheduling module that includes successive evolutions and improvements of the policy.OAR is used to schedule jobs both on the CiGri (Grenoble region) and Grid50000 (France) grids. CiGri is a production grid that federates about 500 heterogeneous resources of various Grenoble laboratories to perform computations in physics. MOAIS has also developed the distributed authentication for access to Grid5000.

FlowVR

We introduced in FlowVR a hierarchical component model for the description of the application. This hierarchy of components strongly enforces the modularity of applications and eases debugging, maintenance and application development. The hierarchy is processed before to start the application by traversing the description and applying a sequence of controllers. The three main controllers take care of building the component content, mapping components on the target architecture hosts,and extracting the low level description required to execute the application.

2009 was also a year of consolidation for FlowVR. Part of the code was deeply reorganized. We fixed numerous bugs, improved the installation procedure and reduced the number of dependencies. The latest version, FlowVR 1.7 was released in December 2009.