<div dir="ltr"><br><br><div class="gmail_quote"><div dir="ltr">On Thu, Jul 26, 2018 at 12:16 PM Fande Kong <<a href="mailto:fdkong.jd@gmail.com">fdkong.jd@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Jul 26, 2018 at 9:51 AM, Junchao Zhang <span dir="ltr"><<a href="mailto:jczhang@mcs.anl.gov" target="_blank">jczhang@mcs.anl.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div dir="ltr">Hi, Pierre,<div> From your log_view files, I see you did strong scaling. You used 4X more cores, but the execution time only dropped from 3.9143e+04 to 1.6910e+04.</div><div> From my previous analysis of a GAMG weak scaling test, it looks communication is one of the reasons that caused poor scaling. In your case, VecScatterEnd time was doubled from 1.5575e+03 to 3.2413e+03. Its time percent jumped from 1% to 17%. This time can contribute to the big time ratio in MatMultAdd ant MatMultTranspose, misleading you guys thinking there was load-imbalance computation-wise. </div><div> The reason is that I found in the interpolation and restriction phases of gamg, the communication pattern is very bad. Few processes communicate with hundreds of neighbors with message sizes of a few bytes. </div></div></blockquote><div><br></div><div>We may need to truncate interpolation/restriction operators. Also do some aggressive coarsening. </div></div></div></div></blockquote><div><br></div><div>Coarsening rates are super important for AMG methods. GAMG has two mechanisms for this, 1) squaring the graph, which provides a big Boolean jump in coarsening rate, and 2) dropping threshold (0.0 - 0.08 range, negative for keeping 0 weight edges), which decreases coarsening rate as it is increased. Graph squaring is almost always useful on 3D problems. threshold == 0 is probably a good start.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>Unfortunately, GAMG currently does not support. </div><div><br></div><div><br></div><div>Fande,</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>If we can avoid this pattern algorithmically (which I don't know), or find ways with faster communication (which I am working), then we can get better scalability. </div></div><div class="gmail_extra"><br clear="all"><div><div class="m_3858336841608056323gmail-m_3677229278573272207gmail_signature"><div dir="ltr">--Junchao Zhang</div></div></div>
<br><div class="gmail_quote">On Thu, Jul 26, 2018 at 10:02 AM, Pierre Jolivet <span dir="ltr"><<a href="mailto:pierre.jolivet@enseeiht.fr" target="_blank">pierre.jolivet@enseeiht.fr</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div class="m_3858336841608056323gmail-m_3677229278573272207HOEnZb"><div class="m_3858336841608056323gmail-m_3677229278573272207h5"><br>
<br>
> On 26 Jul 2018, at 4:24 PM, Karl Rupp <<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a>> wrote:<br>
> <br>
> Hi Pierre,<br>
> <br>
>> I’m using GAMG on a shifted Laplacian with these options:<br>
>> -st_fieldsplit_pressure_ksp_type preonly<br>
>> -st_fieldsplit_pressure_pc_composite_type additive<br>
>> -st_fieldsplit_pressure_pc_type composite<br>
>> -st_fieldsplit_pressure_sub_0_ksp_pc_type jacobi<br>
>> -st_fieldsplit_pressure_sub_0_pc_type ksp<br>
>> -st_fieldsplit_pressure_sub_1_ksp_pc_gamg_square_graph 10<br>
>> -st_fieldsplit_pressure_sub_1_ksp_pc_type gamg<br>
>> -st_fieldsplit_pressure_sub_1_pc_type ksp<br>
>> and I end up with the following logs on 512 (top) and 2048 (bottom) processes:<br>
>> MatMult 1577790 1.0 3.1967e+03 1.2 4.48e+12 1.6 7.6e+09 5.6e+03 0.0e+00 7 71 75 63 0 7 71 75 63 0 650501<br>
>> MatMultAdd 204786 1.0 1.3412e+02 5.5 1.50e+10 1.7 5.5e+08 2.7e+02 0.0e+00 0 0 5 0 0 0 0 5 0 0 50762<br>
>> MatMultTranspose 204786 1.0 4.6790e+01 4.3 1.50e+10 1.7 5.5e+08 2.7e+02 0.0e+00 0 0 5 0 0 0 0 5 0 0 145505<br>
>> [..]<br>
>> KSPSolve_FS_3 7286 1.0 7.5506e+02 1.0 9.14e+11 1.8 7.3e+09 1.5e+03 2.6e+05 2 14 71 16 34 2 14 71 16 34 539009<br>
>> MatMult 1778795 1.0 3.5511e+03 4.1 1.46e+12 1.9 4.0e+10 2.4e+03 0.0e+00 7 66 75 61 0 7 66 75 61 0 728371<br>
>> MatMultAdd 222360 1.0 2.5904e+0348.0 4.31e+09 1.9 2.4e+09 1.3e+02 0.0e+00 14 0 4 0 0 14 0 4 0 0 2872<br>
>> MatMultTranspose 222360 1.0 1.8736e+03421.8 4.31e+09 1.9 2.4e+09 1.3e+02 0.0e+00 0 0 4 0 0 0 0 4 0 0 3970<br>
>> [..]<br>
>> KSPSolve_FS_3 7412 1.0 2.8939e+03 1.0 2.66e+11 2.1 3.5e+10 6.1e+02 2.7e+05 17 11 67 14 28 17 11 67 14 28 148175<br>
>> MatMultAdd and MatMultTranspose (performed by GAMG) somehow ruin the scalability of the overall solver. The pressure space “only” has 3M unknowns so I’m guessing that’s why GAMG is having a hard time strong scaling. <br>
> <br>
> 3M unknowns divided by 512 processes implies less than 10k unknowns per process. It is not unusual to see strong scaling roll off at this size. Also note that the time per call(!) for "MatMult" is the same for both cases, indicating that your run into a latency-limited regime.<br>
> <br>
> Also, have a look at the time ratios: With 2048 processes, MatMultAdd and MatMultTranspose show a time ratio of 48 and 421, respectively. Maybe one of your MPI ranks is getting a huge workload?<br>
<br>
</div></div>Maybe inside GAMG itself (how could I check this?), but since the timing and ratio of the MatMult look OK and the distribution of the pressure space is the same as the other three fields, I’m guessing this does not come from my global Mat, but I may be wrong.<br>
<span><br>
>> For the other fields, the matrix is somehow distributed nicely, i.e., I don’t want to change the overall distribution of the matrix.<br>
>> Do you have any suggestion to improve the performance of GAMG in that scenario? I had two ideas in mind but please correct me if I’m wrong or if this is not doable:<br>
>> 1) before setting up GAMG, first use a PCTELESCOPE to avoid having too many processes work on this small problem<br>
>> 2) have the sub_0_ and the sub_1_ work on two different nonoverlapping communicators of size PETSC_COMM_WORLD/2, do the solve concurrently, and then sum the solutions (only worth doing because of -pc_composite_type additive). I have no idea if this easily doable with PETSc command line arguments<br>
> <br>
> 1) is the more flexible approach, as you have better control over the system sizes after 'telescoping’.<br>
<br>
</span>Right, but the advantage of 2) is that I wouldn't have one half or more of processes idling and I could overlap the solves of both subpc in the PCCOMPOSITE.<br>
<br>
I’m attaching the -log_view for both runs (I trimmed some options).<br>
<br>
Thanks for your help,<br>
Pierre<br>
<br>
<br><br>
> Best regards,<br>
> Karli<br>
<br>
<br></blockquote></div><br></div>
</blockquote></div><br></div></div>
</blockquote></div></div>