Newsgroups: comp.parallel.mpi,comp.unix.cray
From: thimm@physik.fu-berlin.de (Axel Thimm)
Subject: MPI-2 onesided and handshaking?
Organization: FU Berlin, Fachbereich Physik, Institut fuer Theoretische Physik
Date: Mon, 10 Jun 1996 10:52:29 GMT
Message-ID: <4pgups$b53@fu-berlin.de>

Hello,
the EPCC/CRI implementation of MPI already has some draft (?)
implemetations of one-sided communication to appear in MPI-2.
Using onesided communication, the receiver cannot be sure the data has
already arrived, unless
1) he/she checks the data (if the data itself contains a counter) or 
2) uses request-semaphores with thresholds.
But the MPI-2 specs for the T3D warn, that if using theese semaphores
latency might increase too much.
Is there another way of handshaking?
Using twosided communication after the put operation has completed,
could overtake the put transfer (the put might be buffered) and the
handshake would be wrong.

Ragards,
	Axel Thimm.
--
Axel Thimm <thimm@dirac.physik.fu-berlin.de>
Fachbereich Physik, Freie Universitaet Berlin


