World Community Grid - View Thread - Er? Help? Linux takes WAY longer to complete work!

World Community Grid Forums

Category: Completed Research

Forum: FightAIDS@Home

Thread: Er? Help? Linux takes WAY longer to complete work!

Quick Go »

No member browsing this thread

Thread Status: Active
Total posts in this thread: 13

[ ]

Author

This topic has been viewed 2039 times and has 12 replies

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Er? Help? Linux takes WAY longer to complete work!

OK, I have this spare desktop that is set up to dual boot Windows XP and Gentoo Linux 2007.0. I have both set up to use the BOINC client.

Here are the funky results:


os          time            results   avg work time
Windows XP  0:156:11:22:56  530       25508 seconds
Linux       0:004:12:58:31  4         98077 seconds

Keep in mind that this the exact same computer doing the work.
Why is the Linux client taking SO much longer to churn out work compared to Windows XP?

[Sep 26, 2007 4:23:09 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Hello Surface,
My first guess is that the throttle is not set to 100% on your Linux client.
My second guess is that you may have a background process running on Linux that preempts the CPU.
Lawrence

[Sep 26, 2007 4:37:23 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Thanks for the starting points Lawrence.

I ran a quick test to make sure that the faah module was using 100% CPU... it is. When I run top it is constantly in the %99.x range.

As far as background tasks preempting the CPU... I'm a little muddy on what all is exactly setup on my Linux partition. I did shut down X and ran BOINC for the console for two days, but that did not seem to affect the time. There is some 'preempt' terminology in the kernel config, much of which is over my head (or not explained well).

Anyway, as long as BOINC does not restart the faah module, I can run the client overnight and then check top in the morning to see what the CPU time says vs. how much time has really elapsed. Not sure if that will help much though.

[Sep 26, 2007 5:33:38 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

OK, here is some more info...

Running the CPU benchmarks gives these results:

Linux  Benchmark  WindowsXP
1325   Whetstone  1530
2593   Dhrystone  2812

It seems the things that are running in the background in linux are:
avahi
dbus
hald
crond
(I shut these down before running the benchmarks)
Other things to note are: Linux is running in vesafb console mode (no X), and I'm using udev.

The difference between the benchmarks doesn't seem like it would cause such a big time difference... but I really don't know what the benchmarks are measuring.

I'm guessing that, at this point, I need help tuning my kernel.
Unless there ARE some differences between the two BOINC clients?

[Sep 26, 2007 10:11:07 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Hello Surface,
Could you look at the Results Status page and confirm that the last FAAH result took nearly 30 hours to run in Linux? If that is confirmed, then somebody with Linux experience will have to help out.

Lawrence

[Sep 27, 2007 12:13:20 AM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

OS        Status   Sent Date           Return Date         CPU   Points
WindowsXP Valid    09/27/2007 02:14:11 09/27/2007 12:51:04  5.27  48.0 / 51.4
WindowsXP Valid    09/18/2007 08:26:30 09/26/2007 18:44:06  6.97  64.5 / 65.5
Linux     Invalid  09/24/2007 15:19:59 09/25/2007 18:56:14 16.80 103.0 / 21.1
Linux     Valid    09/23/2007 02:35:22 09/24/2007 21:50:43 35.46 246.3 / 53.9

[Sep 27, 2007 2:14:47 PM]

Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Suggest to check what the actual CPU mhz/ghz speed is when the BOINC process is running under Linux. That OS should in fact, so is the word, process a job slightly faster.

----------------------------------------

WCG

Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!

[Sep 27, 2007 2:26:39 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Hello Surface,
The Results Status show that you run your Windows work units at about the same speed as other Windows computers, but that you run your Linux work units much slower than other Linux users. So the problem is not with Linux but with something on your computer. As long as your throttle is at 100%, the only cause that I can conceive is that some other process is running in the background.
Lawrence

[Sep 27, 2007 4:14:58 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Sekerob:

Linux:
Processor: 1 GenuineIntel Intel(R) Pentium(R) 4 CPU 2.80GHz [Family 15 Model 2 Stepping 9][fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr]
 
Windows XP:
Processor: 1 GenuineIntel Intel(R) Pentium(R) 4 CPU 2.80GHz [x86 Family 15 Model 2 Stepping 9]
Processor features: fpu tsc sse sse2 mmx

Lawrence: Here is a ps dump...

Thu Sep 27 17:28:19 CDT 2007
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root         1  0.0  0.1   1532   528 ?        Ss   10:18   0:00 init [3]  
root         2  0.0  0.0      0     0 ?        S<   10:18   0:00 [kthreadd]
root         3  0.0  0.0      0     0 ?        SN   10:18   0:00 [ksoftirqd/0]
root         4  0.0  0.0      0     0 ?        S<   10:18   0:00 [events/0]
root         5  0.0  0.0      0     0 ?        S<   10:18   0:00 [khelper]
root        47  0.0  0.0      0     0 ?        S<   10:18   0:00 [kblockd/0]
root        48  0.0  0.0      0     0 ?        S<   10:18   0:00 [kacpid]
root        49  0.0  0.0      0     0 ?        S<   10:18   0:00 [kacpi_notify]
root       113  0.0  0.0      0     0 ?        S<   10:18   0:00 [ata/0]
root       114  0.0  0.0      0     0 ?        S<   10:18   0:00 [ata_aux]
root       115  0.0  0.0      0     0 ?        S<   10:18   0:00 [kseriod]
root       135  0.0  0.0      0     0 ?        S    10:18   0:00 [pdflush]
root       136  0.0  0.0      0     0 ?        S    10:18   0:00 [pdflush]
root       137  0.0  0.0      0     0 ?        S<   10:18   0:00 [kswapd0]
root       138  0.0  0.0      0     0 ?        S<   10:18   0:00 [aio/0]
root       728  0.0  0.0      0     0 ?        S<   10:18   0:00 [vesafb]
root       810  0.0  0.0      0     0 ?        S<   10:18   0:00 [scsi_eh_0]
root       812  0.0  0.0      0     0 ?        S<   10:18   0:00 [scsi_eh_1]
root       826  0.0  0.0      0     0 ?        S<   10:18   0:00 [kpsmoused]
root       833  0.0  0.0      0     0 ?        S<   10:18   0:00 [kondemand/0]
root       837  0.0  0.0      0     0 ?        S<   10:18   0:00 [reiserfs/0]
root       931  0.0  0.1   1776   608 ?        S<s  10:18   0:00 /sbin/udevd --daemon
root      2054  0.0  0.0      0     0 ?        S<   10:18   0:00 [ksuspend_usbd]
root      2055  0.0  0.0      0     0 ?        S<   10:18   0:00 [khubd]
root      3952  0.0  0.1   1656   564 ?        Ss   10:18   0:00 metalog [MASTER]                                            
root      3953  0.0  0.0   1640   224 ?        S    10:18   0:00 metalog [KERNEL]                                            
root      4424  0.0  0.0   1544   252 ?        Ss   10:18   0:00 /sbin/dhcpcd -h DasGerman eth0
root      4884  0.0  0.2   2340  1136 tty1     Ss   10:18   0:00 /bin/login --     
root      4885  0.0  0.2   2344  1300 tty2     Ss   10:18   0:00 /bin/login --        
root      4886  0.0  0.0   1524   504 tty3     Ss+  10:18   0:00 /sbin/agetty 38400 tty3 linux
root      4887  0.0  0.0   1524   504 tty4     Ss+  10:18   0:00 /sbin/agetty 38400 tty4 linux
root      4888  0.0  0.0   1528   508 tty5     Ss+  10:18   0:00 /sbin/agetty 38400 tty5 linux
root      4889  0.0  0.0   1528   508 tty6     Ss+  10:18   0:00 /sbin/agetty 38400 tty6 linux
root      5776  0.0  0.3   2736  1572 tty1     S    15:49   0:00 -bash
surface   5781  0.0  0.3   2868  1552 tty2     S    15:49   0:00 -bash
surface   6048  0.0  0.6   4976  3468 tty2     S+   15:56   0:04 ./boinc
surface   6049 99.9 15.6 312116 80516 tty2     SNl+ 15:56  91:36 wcg_faah_autodock_5.41_i686-pc-linux-gnu -dpf faah2374_ZINC01653178_xmd01800_00.dpf -gpf ZINC01653178_xmd01800_00.gpf
root      6055  0.0  0.1   2152   864 tty1     R+   17:28   0:00 ps aux

The thing that makes me think that the its not background tasks is that the CPU time is quite high. If it was a background task taking over, the CPU time for FAAH would still be around 5hrs, but BOINC is reporting that the task had control of the CPU for 16-30 hours.

[Sep 28, 2007 1:52:55 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Er? Help? Linux takes WAY longer to complete work!

Hi Surface.

One invalid, and one wildly overrunning and overclaiming task. It's not looking good for Linux, is it?

My colleagues may have been chasing red herrings, since the throttle doesn't affect CPU time - it just makes the task run longer in wall clock time. The same is true of other processes leaching CPU time from WCG.

So, what can cause discrepancies in CPU time? Actually, more than you might expect. Power saving features on the CPU can cause it to underclock. Memory problems have a significant effect, particularly if your L2 cache is damaged or disabled. Finally, intensive IO will affect crunching. Don't run BOINC at the same time as your backup process, and be wary of inconsiderate virus scanners.

[Sep 28, 2007 2:52:33 PM]

[ ]