<html><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">Another option to consider is an option to "bind" the various information (rank, communicator/window information, and perhaps target offset) in a separate step, at least as an option (and implementable as a macro, permitting the compiler to further optimize). For certain applications (those with repetitive access to the same remote process), this would allow separating the process of decoding the "MPI" information and placing it into the bound object. Naturally, this bound object would contain the data needed, eliminating the need for a separate lookup and potential cache miss. There are reasons to consider this at the Exascale as well, even for point-to-point. A sort of "persistent" RMA handle.<div><br></div><div>Bill</div><div><br><div><div>On Sep 1, 2009, at 11:14 AM, Vinod tipparaju wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><span class="Apple-style-span" style="border-collapse: separate; color: rgb(0, 0, 0); font-family: Helvetica; font-size: medium; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: 2; text-align: auto; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: 0px; -webkit-text-decorations-in-effect: none; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; "><span class="Apple-style-span" style="font-family: Verdana; font-size: 13px; "><div><br class="Apple-interchange-newline">You are correct in trying to look at the best possible case and estimating cache-misses/performance-bottlenecks. However, personally don't see any difference between this and shmem. When you cannot really allocate symmetric memory underneath, the amount of bookkeeping is same in both cases. When there is no heterogeneity, the check for this can be disabled at MPI startup. When there is heterogeneity we cannot compare with shmem.</div><div><br></div>I cannot imagine not having symmetric/collective memory object creation to support these RMA interfaces, I think it is a must-have. Sorry I have only been saying we should have these interfaces but haven't given any example for this yet. Given how many times this same issue is coming up, I will do it now.<div><br></div><div>Consider the creation interfaces:</div><div>Create_memobj(IN user_ptr, IN size, OUT mem_obj)</div><div>Create_memobj_collective(user_ptr, size, OUT mem_obj)</div><div>Assign_memobj(IN/OUT mem_obj, IN user_address, IN size) <div><br></div><div>There will be more details on how a mem object which is a result of create_memobj on process A will be exchanged with process B. When it is exchanged explicitly, the heterogeneity information can be created at process B. </div><div><br></div><div>Now take the example with symmetric object:</div><div><br></div><div>Process A </div><div><br></div><div>myptr = allocate(mysize);</div><div>Create_memobj_collective(myptr,mysize, all_obj);<br></div><div>Do all kinds of RMA_Xfers<br></div><div><br></div><div>and an example without symmetric object:</div><div><br></div><div>myptr = allocate(mysize);</div><div>Create_memobj(myptr,mysize,my_obj);</div><div> ----exchange objects here----</div><div>do all kinds of RAM_Xfers</div><div><div><br></div><div>In both cases, I can see being able to communicate without any cache misses for mem_obj.</div><div><br><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">Vinod Tipparaju ^<span class="Apple-converted-space"> </span><a href="http://ft.ornl.gov/~vinod">http://ft.ornl.gov/~vinod</a><span class="Apple-converted-space"> </span>^ 1-865-241-1802</span><br><br><br><br><hr id="stopSpelling">From:<span class="Apple-converted-space"> </span><a href="mailto:keith.d.underwood@intel.com">keith.d.underwood@intel.com</a><br>To:<span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma@lists.mpi-forum.org">mpi3-rma@lists.mpi-forum.org</a><br>Date: Tue, 1 Sep 2009 09:07:41 -0600<br>Subject: Re: [Mpi3-rma] MPI3 RMA Design Goals<br><br><div class="EC_Section1"><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125); ">If we take the SINGLE_RMA_INTERFACE_DRAFT_PROPOSAL as an example, and combine it with the draft design goal #1:<span class="Apple-converted-space"> </span></span>In order to support RMA to arbitrary locations, no constraints on memory, such as symmetric allocation or collective window creation, can be required</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">We get an interesting view on how difficult it can be to get “close to the metal”. So, for MPI_RMA_xfer, we have to assume that the user has some array of target_mem data items. That means the sequence of steps in user space is:</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">target_mem = ranks[dest];</div><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">MPI_RMA_xfer(… target_mem, dest…);</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">If we assume that the message sizes are small and the destinations randomly selected and the machine is large… every access to ranks is a cache miss, and we cannot prevent that by providing fancy hardware. This actually leads me to believe that we may need to reconsider design goal #1, or at least clarify what it means in a way that makes the access more efficient.</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">MPI_RMA_xfer itself is no picnic either. If we take the draft design goal #5: The RMA model must support non-cache-coherent and heterogeneous environments, then MPI is required to maintain a data structure for every rank (ok, it has to do this anyway, but we are trying to get close to the metal) and do a lookup into that data structure with every MPI_RMA_xfer to find out if the target is heterogeneous relative to the target rank – another cache miss. Now, nominally, since this is inside MPI, a lower layer could absorb that check… or, a given MPI could refuse to support heterogeneity or… but, you get the idea. </div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">So, we’ve got two cache line loads for every transfer. One in the application and one in the MPI library. One is impossible to move to the hardware and the other is simply very difficult to move. </div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">For a contrast, look at SHMEM. Assume homogeneous, only one communicator context, and hardware mapping of ranks to physical locations. A shmem_put() of a short item could literally be turned into a few instructions and a processor store (on machines that supported such things). Personally, I think we will have done well if we can get to the point that a reasonable hardware implementation can get MPI RMA to within 2x of a reasonable SHMEM implementation. I think we have a long way to go to get there.</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; ">Keith</div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125); "> </span></p><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125); "> </span></p><div style="border-top-style: none; border-right-style: none; border-bottom-style: none; border-width: initial; border-color: initial; border-left-style: solid; border-left-color: blue; border-left-width: 1.5pt; padding-top: 0in; padding-right: 0in; padding-bottom: 0in; padding-left: 4pt; "><div><div style="border-right-style: none; border-bottom-style: none; border-left-style: none; border-width: initial; border-color: initial; border-top-style: solid; border-top-color: rgb(181, 196, 223); border-top-width: 1pt; padding-top: 3pt; padding-right: 0in; padding-bottom: 0in; padding-left: 0in; "><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><b><span style="font-size: 10pt; font-family: Tahoma, sans-serif; ">From:</span></b><span style="font-size: 10pt; font-family: Tahoma, sans-serif; "><span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma-bounces@lists.mpi-forum.org">mpi3-rma-bounces@lists.mpi-forum.org</a><span class="Apple-converted-space"> </span>[<a href="mailto:mpi3-rma-bounces@lists.mpi-forum.org">mailto:mpi3-rma-bounces@lists.mpi-forum.org</a>]<span class="Apple-converted-space"> </span><b>On Behalf Of<span class="Apple-converted-space"> </span></b>Vinod tipparaju<br><b>Sent:</b><span class="Apple-converted-space"> </span>Tuesday, September 01, 2009 5:23 AM<br><b>To:</b><span class="Apple-converted-space"> </span>MPI 3.0 Remote Memory Access working group<br><b>Subject:</b><span class="Apple-converted-space"> </span>Re: [Mpi3-rma] MPI3 RMA Design Goals</span></div></div></div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "> </p><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">Good points! RMA interfaces should do nothing to prevent utilizing a high message rate (or low overhead communication) that the underlying hardware may offer. To ensure this happens, there should always be a unrestricted path (lets call it this for now, people have called it a "thin layer", "direct access") to the network. </span></div><div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; "> </span></p></div><div><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">This means, despite the fact the the RMA interface has features that abstract out complexity by providing useful characteristics such as ordering and atomicity, it (the RMA interface) should always have this unrestricted, close to the heart of the hardware, path. To achieve this, the unrestricted path should not require any book keeping (from implementation perspective) in relation to the feature-rich path or vice-versa. </span></div></div><div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; "> </span></p></div><div><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">I believe this is what we have demonstrated with the example interfaces hence the null set isn't the case here :-). I will distribute an example implementation very soon so people can get a feel.</span></div></div><div><p class="EC_MsoNormal" style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; "> </span></p></div><div><div><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">---</span></div></div><div><div style="margin-top: 0px; margin-right: 0px; margin-bottom: 0px; margin-left: 0px; padding-top: 0px; padding-right: 0px; padding-bottom: 0px; padding-left: 0px; "><span style="font-size: 10pt; font-family: Verdana, sans-serif; ">Vinod Tipparaju ^<span class="Apple-converted-space"> </span><a href="http://ft.ornl.gov/~vinod">http://ft.ornl.gov/~vinod</a><span class="Apple-converted-space"> </span>^ 1-865-241-1802<br><br><br><br>> From:<span class="Apple-converted-space"> </span><a href="mailto:keith.d.underwood@intel.com">keith.d.underwood@intel.com</a><br>> To:<span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma@lists.mpi-forum.org">mpi3-rma@lists.mpi-forum.org</a><br>> Date: Mon, 31 Aug 2009 16:17:28 -0600<br>> Subject: Re: [Mpi3-rma] MPI3 RMA Design Goals<br>><span class="Apple-converted-space"> </span><br>> There has been stunning silence since this email, so I will go ahead and toss out a thought...<br>><span class="Apple-converted-space"> </span><br>> In the draft design goals, I don't see two issues that I see as key. The first is "support for high message rate/low overhead communications to random targets". As best I can tell, this is one of the key places were the existing one-sided operations are perceived as falling down for existing customers of SHMEM/PGAS. The second is "elimination of the access epoch requirement". This one may be, um, more controversial, but I believe it is part and parcel with the first one. That is, the first one is not that valuable if the programming model requires an excessive amount of access epoch opens and closes just to force the global visibility of the operations. Unfortunately, the intersection of this solution space with the solution space for the current draft design goal #5 (support non-cache-coherent and heterogeneous environments) may be the null set... I will hold out hope that this isn't the case ;-)<br>><span class="Apple-converted-space"> </span><br>> Keith<span class="Apple-converted-space"> </span><br>><span class="Apple-converted-space"> </span><br>> > -----Original Message-----<br>> > From: <a href="mailto:mpi3-rma-bounces@lists.mpi-forum.org">mpi3-rma-bounces@lists.mpi-forum.org</a> [<a href="mailto:mpi3-rma-">mailto:mpi3-rma-</a><br>> ><span class="Apple-converted-space"> </span><a href="mailto:bounces@lists.mpi-forum.org">bounces@lists.mpi-forum.org</a>] On Behalf Of William Gropp<br>> > Sent: Wednesday, August 05, 2009 12:37 PM<br>> > To:<span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma@lists.mpi-forum.org">mpi3-rma@lists.mpi-forum.org</a><br>> > Subject: [Mpi3-rma] MPI3 RMA Design Goals<br>> ><span class="Apple-converted-space"> </span><br>> > I've added versions of the RMA design goals that we discussed at the<br>> > Forum meeting last week to the wiki page for our group (<br>> ><span class="Apple-converted-space"> </span><a href="https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/RmaWikiPage">https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/RmaWikiPage</a><br>> > ). This is a draft; lets discuss these. Also, feel free to add to<br>> > the discussion, particularly in the background section.<br>> ><span class="Apple-converted-space"> </span><br>> > Bill<br>> ><span class="Apple-converted-space"> </span><br>> > William Gropp<br>> > Deputy Director for Research<br>> > Institute for Advanced Computing Applications and Technologies<br>> > Paul and Cynthia Saylor Professor of Computer Science<br>> > University of Illinois Urbana-Champaign<br>> ><span class="Apple-converted-space"> </span><br>> ><span class="Apple-converted-space"> </span><br>> ><span class="Apple-converted-space"> </span><br>> ><span class="Apple-converted-space"> </span><br>> > _______________________________________________<br>> > mpi3-rma mailing list<br>> ><span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma@lists.mpi-forum.org">mpi3-rma@lists.mpi-forum.org</a><br>> ><span class="Apple-converted-space"> </span><a href="http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma">http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma</a><br>><span class="Apple-converted-space"> </span><br>> _______________________________________________<br>> mpi3-rma mailing list<br>><span class="Apple-converted-space"> </span><a href="mailto:mpi3-rma@lists.mpi-forum.org">mpi3-rma@lists.mpi-forum.org</a><br>><span class="Apple-converted-space"> </span><a href="http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma">http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma</a></span></div></div></div></div></div></div></div><span><ATT00001.txt></span></div></span></span></blockquote></div><br><div> <span class="Apple-style-span" style="border-collapse: separate; color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: 2; text-align: auto; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: 0px; -webkit-text-decorations-in-effect: none; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0; "><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><span class="Apple-style-span" style="border-collapse: separate; -webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: 0px; color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; -webkit-text-decorations-in-effect: none; text-indent: 0px; -webkit-text-size-adjust: auto; text-transform: none; orphans: 2; white-space: normal; widows: 2; word-spacing: 0px; "><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div>William Gropp</div><div>Deputy Director for Research</div><div>Institute for Advanced Computing Applications and Technologies</div><div>Paul and Cynthia Saylor Professor of Computer Science</div><div>University of Illinois Urbana-Champaign</div><div><br class="khtml-block-placeholder"></div></div><br class="Apple-interchange-newline"></span></div></span><br class="Apple-interchange-newline"> </div><br></div></body></html>