Joining Team Ukraine to show support?

[Icecold]

I am finishing some tasks or paused ones with deadlines after the Sprint on some of the PG work. I'm a fan of Universe so I'll be in on that. Had to fire up the WSL setup in W10.

Same here, I was running Universe before the Primegrid challenge, so I'm going back to it once the Primegrid challenge is over and will participate in the sprint. In the mean time I'm trying to look around for machines I can fire up to add towards Universe.

 

[SystemVipers]

nice/Ice

 

[mmonnin03]

I didn't know that WSL could help with DC.

Universe runs so much faster in Linux. So for any Windows PCs its worth the effort to setup a VM or use WSL to install a Linux distro. My main PC is W10 so most of work on that PC is native W10 tasks. Universe is one project I'll start up WSL to run Linux tasks.

Double your task throughput on Linux Universe@Home

universeathome.pl

 

[SystemVipers]

Universe runs so much faster in Linux. So for any Windows PCs its worth the effort to setup a VM or use WSL to install a Linux distro. My main PC is W10 so most of work on that PC is native W10 tasks. Universe is one project I'll start up WSL to run Linux tasks.

Yea I was wondering why my Linux box was doing better then my other rigs, didn't even think of that.
All my rigs used to be dual boot, hmmmmm

thanks
SV

 

[mmonnin03]

The thread mentions Ubuntu 20.04 but earlier versions are faster too. Keith mentioned glibc being a reason and 20.04 can have later versions 2.29+. Another reason to update from 18.04 I guess, Universe and SRBase.

Double your task throughput on Linux Universe@Home

universeathome.pl

 

[crashtech]

Universe runs so much faster in Linux. So for any Windows PCs its worth the effort to setup a VM or use WSL to install a Linux distro. My main PC is W10 so most of work on that PC is native W10 tasks. Universe is one project I'll start up WSL to run Linux tasks.

Double your task throughput on Linux Universe@Home

universeathome.pl

Yes, it's known that Linux is faster for many DC projects but I thought I had already tested a previous version of WSL and did not find a performance increase. So, when the PG challenge is over I will set up WSL again and see what the task completion times look like.

 

[StefanR5R]

I was running Universe before the Primegrid challenge, so I'm going back to it once the Primegrid challenge is over and will participate in the sprint.

Same here, I was running U@h for Ukraine before the PG challenge and planned to resume it after it (unless CPDN sends work), and I'll follow that plan even though I don't care for Formula Boinc anymore. Also despite this last bit, I'll modify my plan insofar as I'll run more computers than I need to heat the home.

The thread mentions Ubuntu 20.04 but earlier versions are faster too. Keith mentioned glibc being a reason and 20.04 can have later versions 2.29+.

It most certainly was this change in libmath in glibc release 2.29 which boosted BHSpin and ULX. Since glibc is somewhat modular, it can be sufficient to install a newer libm and libmvec in case of Linux systems which still are on a glibc older than 2.29.

Makes you wonder which other older statically linked DC science application binaries are out there which exercise the same code path, and would speed up simply by relinking them.

 

[mmonnin03]

Yes, it's known that Linux is faster for many DC projects but I thought I had already tested a previous version of WSL and did not find a performance increase. So, when the PG challenge is over I will set up WSL again and see what the task completion times look like.

Linux in WSL isn't faster than a native Linux install, it just allows a Linux OS in Windows. w/o a dual boot.

 

[Skillz]

I was running Universe@Home a week or so ago to get myself to 100M. I don't plan to run it again, but I got a few tasks that should validate for the next few days.

 

[crashtech]

Linux in WSL isn't faster than a native Linux install, it just allows a Linux OS in Windows. w/o a dual boot.

What I did not see in my previous testing was running Universe in a Linux VM within WSL being faster than Universe in that same Windows install. It could be that I was doing something wrong.

 

[StefanR5R]

Current standings from FB sprint 2 stats, league 1 tab:
Ukraine finished in 4th on day one.
Today on day two of three, Ukraine is still in 3rd but had the 2nd best output in the last few hours. As result validations proceed, this may still improve. :-)

 

[Icecold]

Linux in WSL is definitely way faster for Universe than running natively. I let a handful of tasks run overnight in Windows on a machine with a 3950x. (most of my machines are Linux, but a few are Windows). They all took around 14,000 seconds. I installed WSL today (running Ubuntu through WSL), and the tasks are taking around 5,200 seconds on the same exact machine. With that said, another 3950x I have running Linux natively is taking around 3200 seconds for tasks. I'm not sure off hand though if both are set to the same eco mode settings, what the RAM speed is, etc. so it may not be apples to apples.

 

[mmonnin03]

Universe makes it hard to compare apples to apples. Some tasks take 20-21min, some take 1hr on the same system. Min and max run times are def faster in Linux, either native Linux or WSL

 

[StefanR5R]

It looks like benchmarking BHspin with a fixed workunit could be easy: Copy the file "param.in" from the slots directory of a running task into a scratch directory (or copy a workunit input file from projects/universeathome.pl_universe/ into a scratch directory and name this file "param.in"), then launch projects/universeathome.pl_universe/BHspin2_20_x86_64-pc-linux-gnu within this directory. Launch several instances of this at once in parallel in order to create a multicore workload. (Here is example code for launching and timing parallel jobs.) Then either measure the time it takes for a full run, or to make the test quicker, until a desired number of checkpoints were written.

 

[SystemVipers]

hey just got in, looked @FB and Team Ukraine is rocking.
And PG 321 WOW!

feeling good!
Talk tomorrow
SV

 

[StefanR5R]

It looks like benchmarking BHspin with a fixed workunit could be easy:

I kicked the following smoke test off last night: On my desktop Xeon E3, 4-core Haswell, I left 2 threads running Universe in BOINC, 1 thread in this makeshift test with one and the same workunit 3 times in succession (i.e. serial, not parallel), and more or less background activity on the desktop:
for x in 1 2 3; do time /var/lib/boinc/projects/universeathome.pl_universe/BHspin2_20_x86_64-pc-linux-gnu; done

Output:

real 74m37.490s
user 74m35.498s
sys 0m0.189s

real 74m48.501s
user 74m46.561s
sys 0m0.125s

real 74m52.149s
user 74m50.289s
sys 0m0.063s​

These are extremely consistent run times, considering the random nature of the rest of the system load. (Edit: And it is he same run time as when I ran this workunit inside boinc.) After I started this, I also picked out another — hopefully considerably smaller — workunit from BOINC on this computer, and might turn this into a little benchmarking script. Or better yet, into a Linux script and a Windows script. After all, whenever there is new gear appearing, repeatable DC related benchmarks are in demand.

[ Zen 2 for Distributed Computing: Any interest? — Zen 2, clock speed vs power consumption vs ppd sweet spot? — New 5950x DC box, linux 5.13 kernel, mint 20.3 — EPYC 7542 vs Ryzen 5950x in WCG — 12700k vs 5900x/3900x/5950x DC benchmarks info, and some build info. — Zen 3 processors with 3x the L3 cache: Impact on DC projects ]

 

[Icecold]

So what happens if FB can never get a stats update before the end? The points stay as they are currently?

 

[StefanR5R]

I seem to remember one case in the past in which the stats remained stuck at a point several hours before the planned end of the sprint, and another case in which the final update went through half an hour or more after the planned end. But I don't remember which sprints these were.

One would think that FB's stats fetching is more robust this year, as there are fewer teams registered. But evidently, it's not.

 

[Skillz]

Pretty sure Seb cancelled those sprints in years past.

 

[StefanR5R]

So there was, astonishingly, a 20:07 UTC update to the FB stats. Though the following teams have "last hour" = "last 3 hours": Ukraine, Canada, Litomysl, Russia, Austria, Canucks.

The outcome is a random one, though at least among the top teams a plausible outcome, although rather close and hence not entirely conclusive between P3D and DPC.

 

[Icecold]

I switched my CPU's to Yafu and GPU's to Moo! Wrapper both under Ukraine.

 

[SystemVipers]

04/07/2022 13:00 (UTC) - 04/10/2022 12:59 (UTC)
Project to be determined



Hey, been dealing with a death in my orbit, very close friends lovely wife.
Haven't felt like posting but since there is another sprint starting Vsoon I
figured I'd post.

Still see a lot of you rocking the Team Ukraine, me too.
wondering what your thoughts are?

regards
SV

Post on XS - https://xs4s.org/index.php?topic=30190.msg96133#msg96133

 

[Icecold]

I'm sorry to hear of your loss Steve.

I'm planning on running the sprint as part of Ukraine, as long as it's a project that can most likely handle a sprint.

 

[Orange Kid]

I'll be running whatever it is for Ukraine.

 

[SystemVipers]

me to