Loading...
Please wait, while we are loading the content...
Similar Documents
High Performance MPI Implementation On A Network Of Workstations
| Content Provider | Semantic Scholar |
|---|---|
| Author | Lauria, Mario |
| Copyright Year | 1996 |
| Abstract | Despite the emergence of fast LANs, the communication performance available to user applications on workstations still falls short of the level of performance seen on MPPs. A radically new generation of messaging layers are needed to take advantage of the hardware performance and to deliver it to the application level. The thesis presented in this work is that is possible to achieve high enough application level performance on a network of workstations to compete with commodity MPPs. The thesis is proved by presenting MPI-FM, a high performance implementation of a industry standard user level library realized on a network of workstation, and by comparing the achieved results with those published for the same library on two MPPs. When compared to MPI-F, one of the fastest incarnations of the standard, MPI-FM outperforms MPI-F in latency and bandwidth for message of 2KB of size or less. MPI-FM is a high performance full implementation of MPI for network of workstations connected with a Myrinet network. Built on top of the Fast Messages (FM) library, MPI-FM achieves a minimum one-way latency of 19 s and a peak bandwidth of 17.3MB/s using common MPI function calls. These results are obtained on common SPARCstation 20/71 workstations, and no special hardware is required besides a relatively inexpensive Myrinet network. Myrinet is a high speed LAN composed of network interfaces connected by 80 MB/s bidirectional links to crossbar switches, interconnected in an arbitrary topology. At time of writing, the improvement over the performance a orded by legacy protocols on workstation clusters is of two orders of magnitude for latency, one order for bandwidth. MPI implementations with comparable performance can be found only on contemporary MPPs. This work describe some of the issues found in building a high level messaging library (MPI) on top of a low level messaging layer (FM), and our solutions to them. One such issue was the iii direct and e cient support of common operations like adding and removing a header. Another was the exchange of critical information, like the address of the destination bu er, between the layers. In short, a closer integration of the di erent components of the communication software stack was the strategy that allowed us a decisive reduction in the software overhead. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www-csag.ucsd.edu/papers/csag/external/lauria-ms.ps |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
