Newsgroups: comp.parallel.mpi
From: joel@ssd.intel.com (Joel Clark)
Subject: Re: T3D MPI Send/Wait (was Re: PVM send buffers)
Organization: Supercomputer Systems Division (SSD), Intel
Date: 8 Apr 1996 16:36:26 GMT
Message-ID: <4kbfaa$806@news.ssd.intel.com>

In article <4k18cq$ie7@murrow.corp.sgi.com> salo@mrjones.engr.sgi.com (Eric Salo) writes:
>> The most straightforward conversion from csend/crecv for a T3d would be
>> to MPI_Send/MPI_Recv.  That is a match for the semantics.  
>
>Rusty, I'm not sure that I agree. Doesn't csend always return even if the
>matching receive has not yet been posted? We've been bitten many times by
>NX ports that used MPI_Send as a drop-in replacement for csend because we
>don't buffer most messages by default. MPI_Bsend is a better match for
>the semantics, I think.
>
>Eric Salo         Silicon Graphics Inc.             "Do you know what the

Eric, csend does not always return immediately.  Runtime switches establish
a receive side buffer (default 1MB). Csend will return quickly even when
no matching receive is posted if the receiver has enough buffer for the
entire message.  Csend does not return until it knows the entire message
has been sent.  This is better the MPI_Bsend in that csend does not always
result in a copy. MPI_Send matches this semantic, it allows buffer if 
neccessary.

joel






