<div dir="ltr"><div class="gmail_default" style="font-family:tahoma,sans-serif;font-size:small;color:#000000">That's great. Thanks for creating this great piece of software!</div><div class="gmail_default" style="font-family:tahoma,sans-serif;font-size:small;color:#000000"><br></div><div class="gmail_default" style="font-family:tahoma,sans-serif;font-size:small;color:#000000">Amin</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 25, 2020 at 5:56 PM Matthew Knepley <<a href="mailto:knepley@gmail.com">knepley@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div dir="ltr">On Wed, Mar 25, 2020 at 5:41 PM Amin Sadeghi <<a href="mailto:aminthefresh@gmail.com" target="_blank">aminthefresh@gmail.com</a>> wrote:<br></div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Junchao, thank you for doing the experiment, I guess TACC Frontera nodes have higher memory bandwidth (maybe more modern CPU architecture, although I'm not familiar as to which hardware affect memory bandwidth) than Compute Canada's Graham. </div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Mark, I did as you suggested. As you suspected, running make streams yielded the same results, indicating that the memory bandwidth saturated at around 8 MPI processes. I ran the experiment on multiple nodes but only requested 8 cores per node, and here is the result:</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">1 node (8 cores total): 17.5s, 6X speedup</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">2 nodes (16 cores total): 13.5s, 7X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">3 nodes (24 cores total): 9.4s, 10X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">4 nodes (32 cores total): 8.3s, 12X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">5 nodes (40 cores total): 7.0s, 14X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small"><font color="#ff0000">6 nodes (48 cores total): 61.4s, 2X speedup [!!!]</font><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">7 nodes (56 cores total): 4.3s, 23X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">8 nodes (64 cores total): 3.7s, 27X speedup<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><b>Note:</b> as you can see, the experiment with 6 nodes showed extremely poor scaling, which I guess was an outlier, maybe due to some connection problem?</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">I also ran another experiment, requesting 2 full nodes, i.e. 64 cores, and here's the result:</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">2 nodes (64 cores total): 6.0s, 16X speedup [32 cores each node]<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">So, it turns out that given a fixed number of cores, i.e. 64 in our case, much better speedups (27X vs. 16X in our case) can be achieved if they are distributed among separate nodes.</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Anyways, I really appreciate all your inputs.</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><b>One final question:</b> From what I understand from Mark's comment, PETSc at the moment is blind to memory hierarchy, is it feasible to make PETSc aware of the inter and intra node communication so that partitioning is done to maximize performance? Or, to put it differently, is this something that PETSc devs have their eyes on for the future?</div></div></blockquote><div><br></div><div>There is already stuff in VecScatter that knows about the memory hierarchy, which Junchao put in. We are actively working on some other node-aware algorithms.</div><div><br></div><div> Thanks,</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Sincerely,</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Amin</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 25, 2020 at 3:51 PM Junchao Zhang <<a href="mailto:junchao.zhang@gmail.com" target="_blank">junchao.zhang@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">I repeated your experiment on one node of TACC Frontera,<div>1 rank: 85.0s</div><div>16 ranks: 8.2s, 10x speedup</div><div>32 ranks: 5.7s, 15x speedup<br><div><br clear="all"><div><div dir="ltr"><div dir="ltr">--Junchao Zhang</div></div></div><br></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 25, 2020 at 1:18 PM Mark Adams <<a href="mailto:mfadams@lbl.gov" target="_blank">mfadams@lbl.gov</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">Also, a better test is see where streams pretty much saturates, then run that many processors per node and do the same test by increasing the nodes. This will tell you how well your network communication is doing.<div><br></div><div>But this result has a lot of stuff in "network communication" that can be further evaluated. The worst thing about this, I would think, is that the partitioning is blind to the memory hierarchy of inter and intra node communication. The next thing to do is run with an initial grid that puts one cell per node and the do uniform refinement, until you have one cell per process (eg, one refinement step using 8 processes per node), partition to get one cell per process, then do uniform refinement to get a reasonable sized local problem. Alas, this is not easy to do, but it is doable.</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 25, 2020 at 2:04 PM Mark Adams <<a href="mailto:mfadams@lbl.gov" target="_blank">mfadams@lbl.gov</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">I would guess that you are saturating the memory bandwidth. After you make PETSc (make all) it will suggest that you test it (make test) and suggest that you run streams (make streams).<div><br></div><div>I see Matt answered but let me add that when you make streams you will seed the memory rate for 1,2,3, ... NP processes. If your machine is decent you should see very good speed up at the beginning and then it will start to saturate. You are seeing about 50% of perfect speedup at 16 process. I would expect that you will see something similar with streams. Without knowing your machine, your results look typical.</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Mar 25, 2020 at 1:05 PM Amin Sadeghi <<a href="mailto:aminthefresh@gmail.com" target="_blank">aminthefresh@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Hi,</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">I ran KSP example 45 on a single node with 32 cores and 125GB memory using 1, 16 and 32 MPI processes. Here's a comparison of the time spent during KSP.solve:</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">- 1 MPI process: ~98 sec, speedup: 1X</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">- 16 MPI processes: ~12 sec, speedup: ~8X</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">- 32 MPI processes: ~11 sec, speedup: ~9X<br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Since the problem size is large enough (8M unknowns), I expected a speedup much closer to 32X, rather than 9X. Is this expected? If yes, how can it be improved?</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">I've attached three log files for more details. </div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)"><br></div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Sincerely,</div><div style="font-family:tahoma,sans-serif;font-size:small;color:rgb(0,0,0)">Amin</div></div>
</blockquote></div>
</blockquote></div>
</blockquote></div>
</blockquote></div>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>-- Norbert Wiener</div><div><br></div><div><a href="http://www.cse.buffalo.edu/~knepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br></div></div></div></div></div></div></div></div>
</blockquote></div>