<html>
<head>
<style>
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Verdana
}
</style>
</head>
<body class='hmmessage'>
<div>>My argument is that any RMA depends on a call at the origin being able to trigger activity at the target. Modern RMA hardware has the hooks to do the remote side of<tt> MPI_Fast_RMA_xfer()</tt> efficiently based on a call at the origin. Because these hooks are in the hardware they are simply there. They do not use the CPU or hurt performance of things that do use the CPU.<br></div><div><br></div><div>I read this as an argument that says two interfaces are not necessary. </div><div><br></div><div>Having application author promise (during init) it will not do anything that needs an agent is certainly useful. Particularly when, as you state, "<span class="Apple-style-span" style="font-family: monospace, Verdana, Arial, sans-serif; ">having this agent standing by hurts general performance".</span></div><div><br></div>The things that potentially cannot be done without an agent (technically, everything but atomics could be done with out need for any agents) are users choice through explicit usage. Users choses these attributes being aware of their cost hence they can indicate that they will not use them ahead of time when they don't use them.<div><br></div><div>I have repeatedly considered dropping the atomicity attribute, I am unable to because it makes programming (and thinking) so much easier for many applications.</div><div><br></div><div>Vinod.<br><br><br><hr id="stopSpelling">To: mpi3-rma@lists.mpi-forum.org<br>From: treumann@us.ibm.com<br>Date: Wed, 16 Sep 2009 14:18:15 -0400<br>Subject: Re: [Mpi3-rma] non-contiguous support in RMA & one-sided pack/unpack (?)<br><br>
The assertion could then be: MPI_NO_SLOW_RMA (also a bit tongue in cheek)<br>
<br>
My argument is that any RMA depends on a call at the origin being able to trigger activity at the target. Modern RMA hardware has the hooks to do the remote side of<tt> MPI_Fast_RMA_xfer()</tt> efficiently based on a call at the origin. Because these hooks are in the hardware they are simply there. They do not use the CPU or hurt performance of things that do use the CPU.<br>
<br>
RMA hardware may not have the hooks to do the target side of any arbitrary <tt>MPI_Slow_RMA_xfer(). </tt><tt>As a result, support for the more complex RMA_xfer may require a wake-able software agent (thread maybe) to be standing by at all tasks just because they may become target of a Slow_RMA_xfer. </tt><br>
<br>
<tt>If having this agent standing by hurts general performance of MPI applications that will never make a call to Slow_RMA_xfer, why not let the applications author promise up front "I have no need of this agent."</tt><br>
<br>
<tt>An MPI implementation that can support Slow_RMA_xfer with no extra costs (send/recv latency, memory, packet interrupts, CPU contention) will simply ignore the assertion.</tt><br>
<br>
<tt>BTW - I just took a look at the broad proposal and it may contain several things that cannot be done without a wake-able remote software agent. That argues for Keith's idea of an RMA operation which closely matches what RMA hardware does and a second one that brings along all the bells and whistles. Maybe the assertion for an application that only uses the basic RMA call or uses no RMA at all could be MPI_NO_KITCHEN_SINK (even more tongue in cheek).</tt><br>
<br>
<tt> Dick </tt><br>
<br>
<br>
Dick Treumann - MPI Team <br>
IBM Systems & Technology Group<br>
Dept X2ZA / MS P963 -- 2455 South Road -- Poughkeepsie, NY 12601<br>
Tele (845) 433-7846 Fax (845) 433-8363<br>
<br>
<br>
<tt>mpi3-rma-bounces@lists.mpi-forum.org wrote on 09/16/2009 01:08:51 PM:<br>
<br>
> [image removed] </tt><br>
<tt>> <br>
> Re: [Mpi3-rma] non-contiguous support in RMA & one-sided pack/unpack (?)</tt><br>
<tt>> <br>
> Underwood, Keith D </tt><br>
<tt>> <br>
> to:</tt><br>
<tt>> <br>
> MPI 3.0 Remote Memory Access working group</tt><br>
<tt>> <br>
> 09/16/2009 01:09 PM</tt><br>
<tt>> <br>
> Sent by:</tt><br>
<tt>> <br>
> mpi3-rma-bounces@lists.mpi-forum.org</tt><br>
<tt>> <br>
> Please respond to "MPI 3.0 Remote Memory Access working group" </tt><br>
<tt>> <br>
> But, going back to Bill’s point: performance across a range of <br>
> platforms is key. While you can’t have a function for every usage <br>
> (well, you can, but it would get cumbersome at some point), it may <br>
> be important to have a few levels of specialization in the API. <br>
> E.g. you could have two variants:</tt><br>
<tt>> </tt><br>
<tt>> MPI_Fast_RMA_xfer(): no data types, no communicators, etc.</tt><br>
<tt>> MPI_Slow_RMA_xfer(): include the kitchen sink.</tt><br>
<tt>> </tt><br>
<tt>> Yes, the naming is a little tongue in cheek ;-)</tt><br>
<tt>> </tt><br>
<tt>> Keith</tt><br>
<tt>> </tt><br>
<tt>> <snip></tt><BR></div></body>
</html>