Message boards :
Number crunching :
Double your task throughput on Linux
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · Next
Author | Message |
---|---|
Send message Joined: 22 May 18 Posts: 4 Credit: 4,414,233 RAC: 0 ![]() |
From eyeballing my results from windows vs ubuntu I reckon that universe@home is 3 times faster on ubuntu and einstein@home is 4.5 times faster on ubuntu. It looks likes rosetta may be faster on windows and the other projects I run seem to be similar on windows and ubuntu. A pity I can't get Roblox (the kids game) to play on ubuntu then. ![]() |
![]() Send message Joined: 4 Apr 15 Posts: 46 Credit: 43,128,567 RAC: 0 ![]() |
From eyeballing my results from windows vs ubuntu I reckon that universe@home is 3 times faster on ubuntu and einstein@home is 4.5 times faster on ubuntu. It looks likes rosetta may be faster on windows and the other projects I run seem to be similar on windows and ubuntu. A pity I can't get Roblox (the kids game) to play on ubuntu then. Could be the overhead Windows imposes on everything as opposed to the much smaller Linux overhead. |
![]() Send message Joined: 4 Apr 15 Posts: 46 Credit: 43,128,567 RAC: 0 ![]() |
From eyeballing my results from windows vs ubuntu I reckon that universe@home is 3 times faster on ubuntu and einstein@home is 4.5 times faster on ubuntu. It looks likes rosetta may be faster on windows and the other projects I run seem to be similar on windows and ubuntu. A pity I can't get Roblox (the kids game) to play on ubuntu then. Could be the overhead Windows imposes on everything as opposed to the much smaller Linux overhead. |
Send message Joined: 1 Nov 17 Posts: 29 Credit: 291,940,933 RAC: 0 ![]() |
Could be the overhead Windows imposes on everything as opposed to the much smaller Linux overhead. I'm not sure this is it, because I also did some tests with Linux Mint before using Ubuntu in Hyper-V and there was a split in task accelleration... some tasks took 3 hours while some just took 1 hour like in Ubuntu: Another possibility is that Virtualbox did something funky with the emulation, but then also only in Mint and not Ubuntu... ![]() |
![]() ![]() Send message Joined: 10 May 20 Posts: 309 Credit: 4,733,484,700 RAC: 0 ![]() |
When comparing Mint to Ubuntu, make sure both are running a glibc library of at least 2.31. We noticed the speedup was specific to the glibc library when running Ubuntu 18.04 with kernel 5.4 and glibc 2.29 against Ubuntu 20.04 with kernel 5.4 and glibc library 2.31. So even with the kernel versions being exactly the same, it was the glibc version that mattered that caused the significant speedup in Universe tasks. ![]() ![]() A proud member of the OFA (Old Farts Association) |
![]() Send message Joined: 9 Sep 17 Posts: 14 Credit: 580,314,930 RAC: 0 ![]() |
Just checked Linux Mint 20.1 (Ubuntu 20.04) with the latest kernel 5.4.0-66-generic x86_64 and it installs glibc 2.31. But when I checked https://www.gnu.org/software/libc/ it says they've released 2.33. The Synaptic Manager does not list glibc so I don't know what package includes it. When I searched for how to get LM to upgrade to glibc 2.33 I didn't find anything but did see a warning not to try it since stuff will break. Is glibc 2.31 sufficient or might glibc 2.33 give better throughput??? ![]() |
![]() ![]() Send message Joined: 10 May 20 Posts: 309 Credit: 4,733,484,700 RAC: 0 ![]() |
No, you can't update the glibc library via any package manager. It is tied too closely to the kernel and other libraries. The main problem is all the symbol libraries have to be updated for EVERY piece of software on the host. Just not realistically doable. The 2.31 glibc library is totally sufficient. I have not noticed any noticeable performance gain for hosts running newer distros with higher glibc versions. Whatever was in the 2.31 glibc library that the Universe apps liked, it still is present in later versions but they produce no significant improvements over 2.31. ![]() ![]() A proud member of the OFA (Old Farts Association) |
![]() Send message Joined: 9 Sep 17 Posts: 14 Credit: 580,314,930 RAC: 0 ![]() |
Thanks Keith, ULX & BHspin are running at speeds similar to reported here. ![]() |
Send message Joined: 18 Feb 17 Posts: 7 Credit: 6,070,100 RAC: 0 ![]() |
Luckily I have a few Linux machines - well, more than a few, but Ubuntu 20.04 has been all kinds of trouble for me. Still, this project is the only one that Windows truly falls flat on. A shame. There used to be some WCG subprojects where Linux faired much better, but now Windows is doing far better on one or two there. Sometimes Linux isn't always the end all, we won't talk about how many issues I've had with Nvidia GPU drivers and why I love my AMD. LHC at home has now conqored me, for now. I don't have the time or inclination to try and troubleshoot each evening after work, much less on weekends. No idea why windows would be a contributing factor - it isn't as though this project demands a ton of resources, unless it is lacking a library of some sort? |
Send message Joined: 21 Jul 18 Posts: 1 Credit: 33,960,071 RAC: 0 ![]() |
I stumbled on this tread when trying to work why my Xeon E5 (non HT) machine was taking longer to do work units than my slower Xeon E3 (with HT) but running Linux. I had suspected that Linux vs Windows was the issue, just looking at results from different wingmen. . Linux almost always won, even against faster Windows machines. So just wanted to say thanks for the confirmation / explanation, and the suggestion of using a VM if I get keen enough. For the Windows guys that want to experiment with Linux, my suggestion is to dig out an old PC and set it up with Linux Mint. First Gen i3 etc can't be given away locally, it costs actual money to take them to the recycle centre. But throw a cheap SSD and another stick of RAM in there, load up Linux Mint, and you have a perfectly usable machine for Internet use. The machines that are too old or not worth repair I use as "organ donors". Then I can strip them out and sell for scrap, better to get $2 at the scrap yard than pay $20 at the transfer station. Anyway, you can get a machine that will run Linux just fine for practically nothing. This is little "project". Started out as a free i3 based slimline business PC that I got for free. Used lower spec Xeons are cheap on AliExpress, so it's worth looking up the system board and seeing what it will support. https://universeathome.pl/universe/show_host_detail.php?hostid=593120 If I switched it to Linux it would get a big boost on this project. |
![]() Send message Joined: 18 Jul 17 Posts: 138 Credit: 1,379,173,617 RAC: 0 ![]() |
https://universeathome.pl/universe/show_host_detail.php?hostid=599526 Hurray! Made it into the top 40 :) Tom M A proud member of the OFA (Old Farts Assoc.) |
![]() Send message Joined: 18 Jul 17 Posts: 138 Credit: 1,379,173,617 RAC: 0 ![]() |
Re-aimed 3950x CPU to mostly do U@h and the rac is rising nicely. A proud member of the OFA (Old Farts Assoc.) |
Send message Joined: 5 Mar 22 Posts: 3 Credit: 36,465,333 RAC: 0 ![]() |
Installed Linux Mint in VirtualBox under Windows 11, works very good, much better runtimes. |
Send message Joined: 15 Jan 17 Posts: 5 Credit: 20,712,233 RAC: 0 ![]() |
Keith: FWIW, on Linux part of glibc is called "libm", a math library, and the Universe app uses math functions from it (quite heavily). Every now & then the math functions in libm get updated with new more optimized versions that are faster. That's why you see improved times with newer glibc versions. Windows probably doesn't bother with providing any optimized versions. |
Send message Joined: 9 Nov 17 Posts: 21 Credit: 563,207,000 RAC: 0 ![]() |
On 25 February 2021 Keith Myers wrote: We noticed the speedup was specific to the glibc library when running Ubuntu 18.04 with kernel 5.4 and glibc 2.29 against Ubuntu 20.04 with kernel 5.4 and glibc library 2.31.Are you sure about the glibc version of your Ubuntu 18.04 test case? According to distrowatch, Ubuntu 18.04 comes with glibc 2.27. Empirically, I didn't narrow the relevant glibc update down closer than >2.27 && ≤2.31 myself. But from reading glibc release notes, it appears to me that glibc 2.29 introduced the change which sped up BHspin v2 and ULX. (2.29 release notes, relevant patchset) |
![]() ![]() Send message Joined: 10 May 20 Posts: 309 Credit: 4,733,484,700 RAC: 0 ![]() |
Correct. Ubuntu 18.04 shipped with glibc version 2.27. I misremembered the correct version number. ![]() ![]() A proud member of the OFA (Old Farts Association) |
![]() Send message Joined: 18 Jul 17 Posts: 138 Credit: 1,379,173,617 RAC: 0 ![]() |
I spoke too soon on another thread. My U@H centered box had locked up for some reason. Then the application wouldn't restart. So un-archived the zip over the top of it and copied in the Pandora BOINC file. Also set the tasks for E@H down to 60 in case the issue was/is an infection from the other system. Crossing fingers. X X ==edit=== Crossing Fingers didn't work. Perhaps bumping the cpu volts to 1.30 will ===edit=== Tom M https://universeathome.pl/universe/show_host_detail.php?hostid=617626 A proud member of the OFA (Old Farts Assoc.) |
![]() Send message Joined: 18 Jul 17 Posts: 138 Credit: 1,379,173,617 RAC: 0 ![]() |
I spoke too soon on another thread. My U@H centered box had locked up for some reason. It is still running. I bumped the cpu threads/cores up to 100% so I could run 30 threads on U@H and let it munch on E@H on a couple of gpus (gtx 1080's) that are living there for the moment. Tom M A proud member of the OFA (Old Farts Assoc.) |
![]() Send message Joined: 18 Jul 17 Posts: 138 Credit: 1,379,173,617 RAC: 0 ![]() |
Has your RAC been going down due to the in-ability to upload due to Pentathlon contest? Mine apparently has. A proud member of the OFA (Old Farts Assoc.) |
Send message Joined: 4 May 21 Posts: 4 Credit: 5,927,311,700 RAC: 0 ![]() |
Yes, very much so :( My machines are idle most of the time due to not being able to upload finished WU's and thus not being able to download new ones. I've been trying to hit update on the project but it doesn't seem to do any difference. I hope this contest will end soon... |