Message boards : Number crunching : Odd validation (Linux Client sometimes 10x faster than Win32 ?)
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile FalconFly
Avatar

Send message
Joined: 2 Sep 04
Posts: 121
Credit: 592,214
RAC: 0
Message 6443 - Posted: 6 Mar 2005, 9:36:48 UTC
Last modified: 6 Mar 2005, 9:38:57 UTC

I've just uploaded some Results, and saw that the Results other Users submitted all took significantly! longer than mine, on various machines with comparable performance.

My Results validate as okay, however.
http://lhcathome.cern.ch/workunit.php?wuid=75861
( Linux @ 3567s vs. 50205s | 36170s | 23384s | 43501s @ Win2000/XP/2003, all Valid)

http://lhcathome.cern.ch/workunit.php?wuid=40368
( Linux @ 3129s vs. 149035s | 31016s @ WinME/XP, all Valid )

That leaves me with 2 possibilities :
- Validator screwed up OR
- Linux Client is sometimes 10x faster on some WorkUnits

I wouldn't mind getting upto 18x more Credits Granted than Claimed (my Linux BOINC Cores are fully optimized, therefor they claim a realistic number), but that "speed advantage" seems waay to big to be "normal" by any means.
Scientific Network : 45000 MHz - 77824 MB - 1970 GB
ID: 6443 · Report as offensive     Reply Quote
Antjest

Send message
Joined: 30 Sep 04
Posts: 21
Credit: 1,442,034
RAC: 0
Message 6450 - Posted: 6 Mar 2005, 17:01:12 UTC

Boinc CC benchmarks are "better" for Win than Linux.

Try getting optimised Linux CC from SETI site, which gives similar benchmarks to Win.

There is also bug in CC with Win9x/ME, that keeps CPU timer running for all projects, even when paused.

Tony
ID: 6450 · Report as offensive     Reply Quote
Profile FalconFly
Avatar

Send message
Joined: 2 Sep 04
Posts: 121
Credit: 592,214
RAC: 0
Message 6453 - Posted: 6 Mar 2005, 17:50:01 UTC - in response to Message 6450.  
Last modified: 6 Mar 2005, 17:50:59 UTC

As I wrote, my BOINC is already fully optimized, Benchmark Results are pretty much equal to Win32 ones.

The actual issue is as to why a machine running Linux either completes the same job more than 10x faster, or why possibly invalid/differing (my ultra-fast) results are labeled Valid.

If I was running the default, unoptimized BOINC Cores, difference in Credits would be almost a Factor of 30-40x...
Scientific Network : 45000 MHz - 77824 MB - 1970 GB
ID: 6453 · Report as offensive     Reply Quote
Jim Harris

Send message
Joined: 2 Sep 04
Posts: 11
Credit: 20,824
RAC: 0
Message 6454 - Posted: 6 Mar 2005, 17:50:09 UTC - in response to Message 6443.  
Last modified: 6 Mar 2005, 17:52:04 UTC

> That leaves me with 2 possibilities :
> - Validator screwed up OR
> - Linux Client is sometimes 10x faster on some WorkUnits
>
> I wouldn't mind getting upto 18x more Credits Granted than Claimed (my Linux
> BOINC Cores are fully optimized, therefor they claim a realistic number), but
> that "speed advantage" seems waay to big to be "normal" by any means.
>

I believe you experienced a 3rd possibility that other users are getting hit with is that cpu time is being reset to 0 and then counts up till workunit completes.

Also there is text in your stderr out:

-core_client_version>4.19-/core_client_version>
-stderr_txt>
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_init() v4.61
boinc_finish(0)

-/stderr_txt>

Other users do not have this text.

I also have an example with a 1,000,000 turn boince workunit:

http://lhcathome.cern.ch/workunit.php?wuid=76170

One user reported with a time of 845.61 seconds and claimed credit of 1.28,
I had a time of 33,748.80 seconds and claimed credit of 99.67.
Granted credit was 1.28 :(

Interestingly this user also has linux os and stderr out text the same as yours.


ID: 6454 · Report as offensive     Reply Quote
Antjest

Send message
Joined: 30 Sep 04
Posts: 21
Credit: 1,442,034
RAC: 0
Message 6456 - Posted: 6 Mar 2005, 19:07:51 UTC - in response to Message 6453.  

> As I wrote, my BOINC is already fully optimized, Benchmark Results are pretty
> much equal to Win32 ones.
>
> The actual issue is as to why a machine running Linux either completes the
> same job more than 10x faster, or why possibly invalid/differing (my
> ultra-fast) results are labeled Valid.
>
> If I was running the default, unoptimized BOINC Cores, difference in Credits
> would be almost a Factor of 30-40x...
>

Perhaps you should try to run Win CC under Wine.

Check this thread http://einstein.phys.uwm.edu/forum_thread.php?id=1208


Tony
ID: 6456 · Report as offensive     Reply Quote
Profile FalconFly
Avatar

Send message
Joined: 2 Sep 04
Posts: 121
Credit: 592,214
RAC: 0
Message 6470 - Posted: 7 Mar 2005, 12:07:48 UTC - in response to Message 6456.  

> I believe you experienced a 3rd possibility that other users are getting hit
> with is that cpu time is being reset to 0 and then counts up till workunit
> completes.

Good point, that actually might be the most likely cause (didn't consider that, since so far I saw only the 0 Credits-Symptom)

> Perhaps you should try to run Win CC under Wine.
>
> Check this thread http://einstein.phys.uwm.edu/forum_thread.php?id=1208

Nah, I'm running Linux because of running Linux; having no piece of Win32 Code was and is my primary goal; installing WINE would not be of benefit to that principle ;)

(only 3 of my Machines are left running Windows; in a few weeks this number will be further reduced to 2)
Scientific Network : 45000 MHz - 77824 MB - 1970 GB
ID: 6470 · Report as offensive     Reply Quote

Message boards : Number crunching : Odd validation (Linux Client sometimes 10x faster than Win32 ?)


©2024 CERN