Re: Cpu performance with higher order Multipole order

From: Volker Springel <vspringel_at_MPA-Garching.MPG.DE>
Date: Fri, 11 Dec 2020 18:51:16 +0100

Hi Tiago,

This is the expected behaviour in many situations when you use the relative tree opening criterion - in this case, the code tries to maintain the same force accuracy independent of the chosen expansion order. Then it can happen that higher order is benefecial in the sense that you need less CPU time (and fewer tree openings) to reach the same force accuracy.

If you instead set TypeOfOpeningCriterion=0, then the opening angle will stay fixed, and selecting higher order will always be slower because you are then bound to do more computations (which however increase the force accuracy).

Regards,
Volker



> On 11. Dec 2020, at 18:17, Tiago Castro <tiagobscastro_at_gmail.com> wrote:
>
> Dear List,
>
> I have run a series of control runs varying the multipole order on Gadget-4. Surprisingly to me, the CPU time significantly reduces with high orders. For MULTIPOLE_ORDER = 2, 3, 4 I have a total CPU time of 22466.12, 14660.34, 12007.28 respectively. The better performance was due to a significant reduction of the time spent in the tree (see the tables below). Shouldn't one expect the computational cost to increase with high order multipoles? I once again appreciate any clarification on this matter.
>
> MO2:
> Step 2367, Time: 1, CPUs: 156, HighestActiveTimeBin: 20
> diff cumulative
> total 78.52 100.0% 22466.12 100.0%
> treegrav 39.11 49.8% 18540.34 82.5%
> treebuild 10.05 12.8% 1321.58 5.9%
> insert 8.90 11.3% 1145.55 5.1%
> branches 0.18 0.2% 22.01 0.1%
> toplevel 0.14 0.2% 33.58 0.1%
> treeforce 29.00 36.9% 17187.79 76.5%
> treewalk 24.09 30.7% 13786.33 61.4%
> treeimbalance 4.60 5.9% 3196.94 14.2%
> treefetch 0.22 0.3% 115.40 0.5%
> treestack 0.09 0.1% 89.12 0.4%
> MO3:
> Step 2399, Time: 1, CPUs: 156, HighestActiveTimeBin: 20
> diff cumulative
> total 72.65 100.0% 14660.34 100.0%
> treegrav 29.72 40.9% 10729.21 73.2%
> treebuild 10.63 14.6% 1340.10 9.1%
> insert 9.23 12.7% 1151.05 7.9%
> branches 0.20 0.3% 24.94 0.2%
> toplevel 0.18 0.2% 47.31 0.3%
> treeforce 19.01 26.2% 9350.97 63.8%
> treewalk 15.91 21.9% 7611.72 51.9%
> treeimbalance 2.49 3.4% 1533.54 10.5%
> treefetch 0.55 0.8% 165.79 1.1%
> treestack 0.06 0.1% 39.92 0.3%
> MO4:
> Step 2415, Time: 1, CPUs: 156, HighestActiveTimeBin: 20
> diff cumulative
> total 73.47 100.0% 12007.28 100.0%
> treegrav 26.37 35.9% 8063.34 67.2%
> treebuild 10.02 13.6% 1386.34 11.5%
> insert 8.93 12.2% 1173.43 9.8%
> branches 0.24 0.3% 31.16 0.3%
> toplevel 0.20 0.3% 65.76 0.5%
> treeforce 16.29 22.2% 6631.74 55.2%
> treewalk 13.57 18.5% 5327.90 44.4%
> treeimbalance 1.85 2.5% 1014.50 8.4%
> treefetch 0.79 1.1% 250.43 2.1%
> treestack 0.08 0.1% 38.91 0.3%
>
> Cheers,
> Tiago Castro Post Doc, Department of Physics / UNITS / OATS
> Phone: (+39 040 3199 120)
> Mobile: (+39 388 794 1562)
> Email: tiagobscastro_at_gmail.com
> Website: tiagobscastro.com
> Skype: tiagobscastro
> Address: Osservatorio Astronomico di Trieste / Villa Bazzoni
> Via Bazzoni, 2, 34143 Trieste TS
>
>
>
> -----------------------------------------------------------
>
> If you wish to unsubscribe from this mailing, send mail to
> minimalist_at_MPA-Garching.MPG.de with a subject of: unsubscribe gadget-list
> A web-archive of this mailing list is available here:
> http://www.mpa-garching.mpg.de/gadget/gadget-list
Received on 2020-12-11 18:51:17

This archive was generated by hypermail 2.3.0 : 2022-09-01 14:03:43 CEST