Abstract | ||
---|---|---|
Interconnecting multiple clusters with a high speed networkto form a single heterogeneous architecture (i.e. acluster of clusters) is currently a hot issue. Consequently,new runtime systems that are able to simultaneously dealwith multiple high speed networks within the same applicationhave to be designed. This paper presents how wedid extend an existing multi-device communication librarywith fast internal data-forwarding capabilities on gatewaynodes. On top of that, efficient high-level routing mechanismscan be implemented. Our approach is easily applicableto many network protocols and is completely in-pendantfrom the application code. Efficiency is achievedby avoiding extra data copies when possible and by usingpipelining techniques. The preliminary experiments showthat the observed inter-cluster bandwidth is close to the onethat can be delivered by the hardware. |
Year | DOI | Venue |
---|---|---|
2001 | 10.1109/IPDPS.2001.925040 | IPDPS |
Field | DocType | ISBN |
Cluster (physics),Architecture,Computer science,Parallel computing,Computer network,Internet protocol suite,Bandwidth (signal processing),Intelligent Network,Distributed computing,Communications protocol | Conference | 0-7695-0990-8 |
Citations | PageRank | References |
3 | 0.52 | 7 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Olivier Aumage | 1 | 160 | 17.10 |
Lionel Eyraud | 2 | 59 | 5.07 |
Raymond Namyst | 3 | 1405 | 83.04 |