| Rusty Lusk, Personal Communication, December 1993. |
....multiple (2 8) CPU s and global memory are becoming commonplace; when collections of such machines are used as a parallel computing platform, the traditional process oriented message passing model is probably suboptimal. Finally, new concepts such as active messages [17] and remote memory copy [11] are attractive but extremely cumbersome to implement within a process oriented framework. In order to investigate the viability of a threads based distributed concurrent computing model to address some of the above issues, we have designed and implemented an experimental system called TPVM ....
Rusty Lusk, Personal Communication, December 1993.
....not address the global variable problem. An internal NEC implementation of MPI, done in Tokyo, for the NEC SX 4 parallel vector machine is the only project we know of that addresses the global variable problem. The compiler (by default) converts all global variables into thread specific data [8]. Since communication is through shared memory, the implementation can then use threads as MPI processes, thereby achieving greater efficiency as we have noted. 3 Threads Only MPI Implementation In this section, we present our threads only MPI implementation for developing parallel programs on a ....
Ewing Lusk, "Personal communication," March 1997.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC