Message boards : Number crunching : 1.6 GHZ takes over 160,000 seconds. Possible to get LHCatH preference for short WU??
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile jay

Send message
Joined: 10 Aug 07
Posts: 54
Credit: 813,704
RAC: 116
Message 26247 - Posted: 8 Mar 2014, 16:29:44 UTC
Last modified: 8 Mar 2014, 16:34:11 UTC

Greetings

This wu took over 163,000 seconds.

http://lhcathomeclassic.cern.ch/sixtrack/workunit.php?wuid=13914849

Yes, it a slower pc - but should it take nearly 5 times as long as a cpu that is twice as fast?
Is the floating processors that slower?
24-Jan-2014 08:22:41 [---] Processor: 2 GenuineIntel Intel(R) Atom(TM) CPU N270 
  @ 1.60GHz [Family 6 Model 28 Stepping 2]
24-Jan-2014 08:22:41 [---] Processor features: fpu vme de tsc msr pae mce cx8 ap
ic sep mtrr pge mca cmov pat clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx 
constant_tsc arch_perfmon pebs bts aperfmperf pni dtes64 monitor ds_cpl est tm2 
ssse3 xtpr pdcm movbe lahf_lm dtherm
24-Jan-2014 08:22:41 [---] OS: Linux: 3.2.0-4-686-pae
24-Jan-2014 08:22:41 [---] Memory: 493.15 MB physical, 953.00 MB virtual
24-Jan-2014 08:22:41 [---] Disk: 4.58 GB total, 531.53 MB free


yes, it has little memory - but I monitored swapping and there was none going on.
It is an HP mini with an ssd drive - there was sufficient drive space as well.

All this said. The other client finished in 32,950. so the WU should be OK.

I have processed other WU on this PC http://lhcathomeclassic.cern.ch/sixtrack/results.php?hostid=10310501 - anther wu took 166,684.99 seconds - over 46 hours.


Main question:
Could an LHCatHome preference be made to avoid long WU??

Thank you all...

Jay

edit - yes, the '6' in the wu name means it takes lots of loops.
But can I avoid these? - end_edit
ID: 26247 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 26248 - Posted: 8 Mar 2014, 16:46:39 UTC - in response to Message 26247.  

Sorry Jay; that was a one million turn task, one of several/many? submitted
by a new user. While some people love these I am afraid our plans for
splitting them are delayed. Maybe the experts can explain if there is a way
to limit who gets them based on the CPU estimate. Will try and follow up.

Eric.
ID: 26248 · Report as offensive     Reply Quote
Richard Haselgrove

Send message
Joined: 27 Oct 07
Posts: 186
Credit: 3,297,640
RAC: 0
Message 26250 - Posted: 9 Mar 2014, 9:21:22 UTC - in response to Message 26248.  

Sorry Jay; that was a one million turn task, one of several/many? submitted
by a new user. While some people love these I am afraid our plans for
splitting them are delayed. Maybe the experts can explain if there is a way
to limit who gets them based on the CPU estimate. Will try and follow up.

Eric.

There is an obscure feature in the BOINC server code called Multi-size apps. It was added about a year ago at the request of WCG:

We recently added server-side support for "multi-size apps":
http://boinc.berkeley.edu/trac/wiki/MultiSize
This lets you create different-size jobs for a given app,
and the scheduler tries to send larger jobs to faster devices.

This feature was requested by WCG,
but I think a number of projects may benefit from it.

Testing status: I've tested this only minimally.
I'd recommend using it in a test project for a while before
moving it to production.

-- David
(26 April 2013)

Everyone seems to have ignored it until last month, when Jon Sonntag of the Collatz Project tried to use it, and found several bugs. It might be getting close to usable now, but I think there are still some unanswered questions.

The recent discussion starts at http://lists.ssl.berkeley.edu/mailman/private/boinc_projects/2014-February/010626.html (boinc_projects mailing list: you have to be a subscriber to access that archive): either Jon or David might be able to help you decide whether the feature will be helpful in your case.
ID: 26250 · Report as offensive     Reply Quote
Profile jay

Send message
Joined: 10 Aug 07
Posts: 54
Credit: 813,704
RAC: 116
Message 26251 - Posted: 9 Mar 2014, 21:42:29 UTC - in response to Message 26250.  

Thank you both!

I don't mind running the longer WU, but on this (slower) PC there is a danger of timing out - especially if I have to shut down while travelling..

Again, thank you for your responses....
Jay
ID: 26251 · Report as offensive     Reply Quote
Profile Ananas

Send message
Joined: 17 Jul 05
Posts: 102
Credit: 542,016
RAC: 0
Message 26261 - Posted: 11 Mar 2014, 6:46:40 UTC
Last modified: 11 Mar 2014, 6:57:38 UTC

A bit OT, but maybe still interesting in this context :

In this one of your results, your wingman has been a (dual CPU) Xeon L5520

The latest Atom CPU (C2750, 8x 2.40GHz) has been benchmarked against the L5520 (single CPU) and those two CPUs have about the same average crunching power - but the L5520 (the "L" indicates it is tagged "low power") has a TDP of 60W whereas the C2750 has only 20W.

So when it comes to Atom CPUs, it is a good idea to compare the efficiency rather than the plain floating point power, because that's what they have been optimized for.
ID: 26261 · Report as offensive     Reply Quote

Message boards : Number crunching : 1.6 GHZ takes over 160,000 seconds. Possible to get LHCatH preference for short WU??


©2024 CERN