View Full Version : DF and kernels
Hey, I thought I'd post something here before my final exams hit and you never hear from me again.
Something to note before you read all the stuff below: these are not well controlled tests nor are they conclusive, I noticed some trends and decide to post because I thought the results were interesting. After reading a thread in the gentoo forums I decided to go and try the 2.5.67 kernel (my computer was getting a bit too complacent ... thought it could start slacking off in its folding duties :jester: ). So after a while I got the idea of comparing the folding capabilities of the 2.5.67 kernel against its 2.4.20 predecessor. So I ran a few tests.
All testing was done using Gentoo linux (http://www.gentoo.org) 1.4rc3. The client used was the regular client (not the beta) and with the current (89aa) protein. All kernels were compiled with a minimum number of extra features enabled (e.g. no framebuffer console, no extra networking options, no unecessary filesystems, no usb), ALSA drivers were used, DRM was enabled, and ReiserFS was used on all systems (mounted with noatime and notail options) DMA was enabled.
A comparison on my sister's computer, usually has X running on two virtual screens, this test was not very well controlled at all:
This is a development-sources 2.5.67 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 15870 structures over 19800 seconds.
Structures Per Second: 0.80
Structures Per Minute: 48.09
Structures Per Hour: 2885.45
Structures Per Day: 69250.80
Linux OS - Running Kernel Version 2.5.67
Intel(R) Pentium(R) III CPU family 1266MHz @ 1272mhz (512 KB cache)
------------------------------------------------------------
This is a gentoo-sources 2.4.19 kernel (not 2.4.20 like mine)
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 15475 structures over 19800 seconds.
Structures Per Second: 0.78
Structures Per Minute: 46.89
Structures Per Hour: 2813.64
Structures Per Day: 67527.36
Linux OS - Running Kernel Version 2.4.19-gentoo-r10
Intel(R) Pentium(R) III CPU family 1266MHz @ 1273mhz (512 KB cache)
------------------------------------------------------------
A 2.55% increase from the older kernel (note that this is the only test where I had the good sense to let cron start the clients so they would be logged at exactly the same time :o ).
Continued in next post due to post size limit (I'm sure you will have lots of fun reading through all this :sleepy: ).
A comparison on my computer with X loaded, running gaim, phoenix, xmms, fluxbox, pan -- I was actively using the computer during this time. The test was done with some control, but the time period was shorter and there may have been some variation:
This is a development-sources 2.5.67 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 13710 structures over 16098 seconds.
Structures Per Second: 0.85
Structures Per Minute: 51.10
Structures Per Hour: 3065.97
Structures Per Day: 73583.28
Linux OS - Running Kernel Version 2.5.67
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
This is a development-sources 2.5.67 kernel (WITH elevator=cfq passed as a boot option)
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 13730 structures over 16606 seconds.
Structures Per Second: 0.83
Structures Per Minute: 49.61
Structures Per Hour: 2976.51
Structures Per Day: 71436.24
Linux OS - Running Kernel Version 2.5.67
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
This is a gentoo-sources 2.4.20 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 12785 structures over 15593 seconds.
Structures Per Second: 0.82
Structures Per Minute: 49.20
Structures Per Hour: 2951.71
Structures Per Day: 70841.04
Linux OS - Running Kernel Version 2.4.20-gentoo-r2
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
Here I was playing with the elevator option in the 2.5 kernel, the improvement with the regular elevator algorithm is 3.87%, with the cfq elevator algorithm it drops to 0.840% (of course this might just be to random variations, again these tests are not well controlled).
A comparison on my computer with no other programs loaded, this test is the best controlled of the lot, the only varation here should be the luck of the folding algorithm, and I managed a longer time period to get a better average:
This is a development-sources 2.5.67 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 22650 structures over 24449 seconds.
Structures Per Second: 0.93
Structures Per Minute: 55.59
Structures Per Hour: 3335.11
Structures Per Day: 80042.64
Linux OS - Running Kernel Version 2.5.67
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
This is a gentoo-sources 2.4.20 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 22640 structures over 25055 seconds.
Structures Per Second: 0.90
Structures Per Minute: 54.22
Structures Per Hour: 3253.00
Structures Per Day: 78072.00
Linux OS - Running Kernel Version 2.4.20-gentoo-r2
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
Here are the processes running for each test above:
This is a gentoo-sources 2.4.20 kernel
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 1 0.0 0.0 1328 480 ? S 08:02 0:04 init
root 2 0.0 0.0 0 0 ? SW 08:02 0:00 [keventd]
root 3 0.0 0.0 0 0 ? SWN 08:02 0:00 [ksoftirqd_CPU0]
root 4 0.0 0.0 0 0 ? SW 08:02 0:00 [kswapd]
root 5 0.0 0.0 0 0 ? SW 08:02 0:00 [bdflush]
root 6 0.0 0.0 0 0 ? SW 08:02 0:00 [kupdated]
root 8 0.0 0.0 0 0 ? SW 08:02 0:00 [kreiserfsd]
root 69 0.0 0.1 1636 844 ? S 08:02 0:00 /sbin/devfsd /dev
root 678 0.0 0.1 1388 544 ? S 08:02 0:00 metalog [MASTER]
root 681 0.0 0.0 1360 452 ? S 08:02 0:00 metalog [KERNEL]
at 682 0.0 0.1 1436 588 ? S 08:02 0:00 [atd]
root 717 0.0 0.1 1488 636 ? S 08:02 0:00 /usr/sbin/fcron
root 884 0.0 0.0 1328 440 ? S 08:02 0:00 /sbin/dhcpcd eth0
root 955 0.0 0.2 2804 1360 ? S 08:02 0:00 /usr/sbin/sshd
root 972 0.0 0.3 3904 1664 ? S 08:02 0:00 /sbin/mount.smbfs
root 1007 0.0 0.1 2220 1012 tty2 S 08:02 0:00 login -- root
root 1008 0.0 0.0 1308 472 tty3 S 08:02 0:00 /sbin/agetty 3840
root 1009 0.0 0.0 1308 472 tty4 S 08:02 0:00 /sbin/agetty 3840
root 1010 0.0 0.0 1308 472 tty5 S 08:02 0:00 /sbin/agetty 3840
root 1011 0.0 0.0 1308 472 tty6 S 08:02 0:00 /sbin/agetty 3840
root 1012 0.0 0.2 2568 1492 tty2 S 08:02 0:00 -bash
root 1018 0.0 0.2 2236 1132 tty1 S 08:03 0:00 [login]
qed 1028 0.0 0.2 2568 1508 vc/1 S 08:05 0:00 -bash
qed 2535 0.0 0.1 2636 740 vc/1 R 15:24 0:00 ps aux
This is a development-sources 2.5.67 kernel
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 1 0.0 0.0 1328 476 ? S Apr11 0:04 init
root 2 0.0 0.0 0 0 ? SWN Apr11 0:00 [ksoftirqd/0]
root 3 0.0 0.0 0 0 ? SW< Apr11 0:00 [events/0]
root 4 0.0 0.0 0 0 ? SW Apr11 0:00 [pdflush]
root 5 0.0 0.0 0 0 ? SW Apr11 0:01 [pdflush]
root 6 0.0 0.0 0 0 ? SW Apr11 0:00 [kswapd0]
root 7 0.0 0.0 0 0 ? SW< Apr11 0:00 [aio/0]
root 8 0.0 0.0 0 0 ? SW Apr11 0:00 [kseriod]
root 9 0.0 0.0 0 0 ? SW< Apr11 0:00 [reiserfs/0]
root 70 0.0 0.1 1664 872 ? S Apr11 0:00 /sbin/devfsd /dev
root 589 0.0 0.1 1412 588 ? S Apr11 0:00 metalog [MASTER]
root 590 0.0 0.0 1360 452 ? S Apr11 0:00 metalog [KERNEL]
at 593 0.0 0.1 1440 580 ? S Apr11 0:00 /usr/sbin/atd
root 628 0.0 0.1 1488 632 ? S Apr11 0:00 /usr/sbin/fcron
root 795 0.0 0.0 1328 444 ? S Apr11 0:00 /sbin/dhcpcd eth0
root 866 0.0 0.2 2804 1356 ? S Apr11 0:00 /usr/sbin/sshd
root 883 0.0 0.3 3968 1720 ? S Apr11 0:00 /sbin/mount.smbfs
root 885 0.0 0.0 0 0 ? SW Apr11 0:00 [smbiod]
root 897 0.0 0.0 1308 472 vc/3 S Apr11 0:00 /sbin/agetty 3840
root 898 0.0 0.0 1308 472 vc/4 S Apr11 0:00 /sbin/agetty 3840
root 899 0.0 0.0 1308 472 vc/5 S Apr11 0:00 /sbin/agetty 3840
root 900 0.0 0.0 1308 472 vc/6 S Apr11 0:00 /sbin/agetty 3840
qed 2432 0.0 0.1 2300 984 ? SN 03:42 0:00 /bin/sh ./foldit
qed 2436 99.8 24.5 134852 126340 ? RN 03:42 609:05 ./foldtrajlite -f
root 2708 0.0 0.2 2240 1132 vc/2 S 05:09 0:00 login -- qed
root 2709 0.0 0.1 2232 1008 vc/1 S 05:09 0:00 login -- root
root 4058 0.0 0.2 2572 1496 vc/1 R 13:48 0:00 -bash
qed 4063 0.0 0.2 2568 1492 vc/2 S 13:48 0:00 -bash
root 4095 0.0 0.1 2640 744 vc/1 R 13:52 0:00 ps aux
Here the new kernel improves on the old by 2.52%, this test might actually be reproducable (I'm going try and do this at some point), but first I'm going to have a look at how the cfq elevator performs under the same conditions.
From my folding logs, no control even attempted, but the time period is quite a bit longer so I am posting it. This is also the data where I first got the idea that 2.5.67 might have a noticable folding improvement over the 2.4.20 kernel:
This is a development-sources 2.5.67 kernel
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 35750 structures over 40389 seconds.
Structures Per Second: 0.89
Structures Per Minute: 53.11
Structures Per Hour: 3186.51
Structures Per Day: 76476.24
Linux OS - Running Kernel Version 2.5.67
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
This is a gentoo-sources 2.4.19 kernel (This was my old kernel, it had lots of stuff compiled into it and so it was not streamlined for the test)
------------------------------------------------------------
Distributed Folding Linux Benchmark Script V1.0
Sample Size: 32885 structures over 40862 seconds.
Structures Per Second: 0.80
Structures Per Minute: 48.29
Structures Per Hour: 2897.22
Structures Per Day: 69533.28
Linux OS - Running Kernel Version 2.4.19-gentoo-r9
AMD Athlon(TM) XP1600+ @ 1410mhz (256 KB cache)
------------------------------------------------------------
Before you look at the numbers and become shocked :scared: ... this was not controlled ... I was quite likely doing something like compiling the 2.5.67 kernel while the 2.4.19 was running :P . Also note that it is a 2.4.19 kernel and not 2.4.20.
I should also note that all of the 2.4 series kernels I used are from the gentoo-sources package in Gentoo, they will obviously have different targets from the raw 2.5.67 kernel or something like wolk-sources or mm-sources. It is entirely possible that selecting a proper set of patches for a 2.4 kernel it could reduce the advantage the 2.5 kernel appears to have. It is also possible that with a proper patch set a 2.4 kernel could outperform a 2.5 kernel (it is a developmental kernel remember). It might be interesting to hear what Chinasaur has to say about this, since in his recent review of Mandrake 9.1 he noted that he was getting 186K per CPU (XP1800 with PC2700), his hardware is better than mine :cry: (XP1600 with PC2100), but I was getting 151K so maybe the Mandrake kernel is making up part of the 35K difference.
I hope this was of interest to some of you, I do think there is a reasonable possibility that the 2.5 kernel will outperform the 2.4 but at this point it is still a lot closer to hypothesis than proven conclusion. If some of you want to run off and install a 2.5 kernel thats great, just remember that things can AND WILL break, and don't be dissappointed if you don't get any performance increase (but please add to this thread -- or at least tell me if no one else is interested). In any case this little review has done wonders for my post count :D.
Happy folding :cheers:
TheOtherPhil
04-13-2003, 06:30 AM
Very interesting...thanks! I run Gentoo on a few boxen so I might have a play when I get the time :)
Chinasaur
04-21-2003, 01:30 PM
Kosh:
:notworthy
:notworthy
:thumbs:
GHOST
06-09-2003, 01:56 AM
On the 157 protien, with a pentium 4 1.6
windows xp gave me 60,000 structures a day.
mandrake 9.1 gave me 70,000 structures a day
redhat 9 gives me 75,000 structures a day
Powered by vBulletin® Version 4.2.4 Copyright © 2024 vBulletin Solutions, Inc. All rights reserved.