New version
log in

Advanced search

Message boards : News : New version

Author Message
Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 592 - Posted: 19 Sep 2012, 7:42:16 UTC
Last modified: 19 Sep 2012, 7:51:59 UTC

I've modified the F77 code to ignore pixel combinations that won't converge.

Hopefully (fingers and toes crossed) this should stop the invalid tasks we've been seeing lately.

I also feel confident enough to start turning on the optimisation so hopefully it will crunch quicker.
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile _aD
Send message
Joined: 16 Sep 12
Posts: 25
Credit: 1,200,733
RAC: 101
Message 595 - Posted: 19 Sep 2012, 16:05:39 UTC - in response to Message 592.

I'm new to BOINC - will the new version be automagically sent to the clients, and if so when can we expect to see it?

Profile Ralf02061973
Avatar
Send message
Joined: 31 Aug 12
Posts: 4
Credit: 5,184,940
RAC: 1,082
Message 596 - Posted: 19 Sep 2012, 16:07:44 UTC - in response to Message 592.
Last modified: 19 Sep 2012, 16:09:32 UTC

hello kevin

i have some questions...

where is this F77 code?
and where are these optimisations?
are they in the wu's?

greetings from germany
ralf


€dit: @_aD ...u was faster ;)
____________
Boinc runs here on:
Intel i7-3770K + IntelHD4000
Android-Stick-ARM-Cotex-A17
Sony-Z5C-ARM-Cortex-A53/A57
Nvidia GT-630 / Nvidia GTX-750Ti

Profile _aD
Send message
Joined: 16 Sep 12
Posts: 25
Credit: 1,200,733
RAC: 101
Message 597 - Posted: 19 Sep 2012, 16:39:57 UTC - in response to Message 596.

Looks like it's all automagical: my 2.10's that are finishing are being replaced by 2.20.

Profile Neil Polson
Avatar
Send message
Joined: 22 Aug 12
Posts: 2
Credit: 100,332
RAC: 0
Message 600 - Posted: 19 Sep 2012, 19:38:48 UTC

The new app doesn't seem to validate against the old one!!
____________

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 601 - Posted: 20 Sep 2012, 1:36:30 UTC - in response to Message 595.

@_aD

Yes it will
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 602 - Posted: 20 Sep 2012, 1:38:48 UTC - in response to Message 596.

@The chosen

The F77 code is called by the BOINC wrapper and does the heavy lifting for us. It is that code that I turned on the optimiser. Once the new version is downloaded you have the optimised executable.
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 603 - Posted: 20 Sep 2012, 1:40:28 UTC - in response to Message 600.

@Neil

It most cases it will validate. It can't validate against an old result that did not converge. The new code stops that happening and should converge many more pixels
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Ray Murray
Avatar
Send message
Joined: 9 Sep 12
Posts: 4
Credit: 39,002
RAC: 0
Message 605 - Posted: 20 Sep 2012, 9:13:55 UTC

Since release of the new version I've had a few very short WUs. Some zipped through in 20 seconds and some have got to c70% very quickly then slowed but still finished in an hour or less rather than the 5 hours I'd expect. 4 very short ones are still pending but this wingman completed in similar time. This one was invalid under 2.10 but valid in a fraction of the time with 2.20.
Don't know if these short WUs are intentional but we seem to be getting the same 164 credits for the 20 second and 1/2 hour work as for 6 hours worth.

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 606 - Posted: 20 Sep 2012, 9:31:47 UTC - in response to Message 605.

@Ray

The short WU's are because the new code says - "this will never converge" - and moves to the next pixel, but being the kind hearted soul I am I give you the credit for processing it.

This will keep happening until the next load of galaxies are loaded and we fix the problem at the source.
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Dingo
Avatar
Send message
Joined: 3 Aug 12
Posts: 121
Credit: 23,101,687
RAC: 5,858
Message 653 - Posted: 27 Sep 2012, 4:01:08 UTC

Looking at most of my Linux work, computer 37, that has a Windows Partner and the Linux PC takes much much longer to do the same work.

Are the "Run Times" numbers just wrong or is the Linux code much slower that the Windows??
____________

Proud Founder and member of



Have a look at my WebCam

Profile Terminal*
Send message
Joined: 1 Sep 12
Posts: 20
Credit: 25,770,587
RAC: 0
Message 654 - Posted: 27 Sep 2012, 4:13:16 UTC - in response to Message 653.

I think the linux time length's are just super wayyyy offf. My 3 Linux machines were getting much faster times when they were on windows, however the average credit/etc has still been going up, so I think they're actually not taking as long as they say.

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 655 - Posted: 27 Sep 2012, 7:46:27 UTC - in response to Message 654.

@Terminal

All I can do is give an estimate of the Gflops - BOINC does the rest
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Kevin
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar
Send message
Joined: 27 Jul 12
Posts: 507
Credit: 14,550,449
RAC: 2,760
Message 656 - Posted: 27 Sep 2012, 8:41:35 UTC - in response to Message 653.

@Dingo

I see it the other way round my Linux machine is ripping through things
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research

Profile Conan
Avatar
Send message
Joined: 16 Aug 12
Posts: 52
Credit: 5,509,244
RAC: 4,833
Message 657 - Posted: 27 Sep 2012, 9:54:01 UTC

Not looking at the run times posted via BOINC in my account, because they don't appear to mean a thing, but looking at the estimated run times on my computers, it would appear that my Linux machines are way faster.

Admittedly my Linux is 64 bit and my Windows is 32 bit and that could also be a difference.

But depending on the WU type it will take between 1.21 Hours and 2.50 hours on my 64bit Linux computers.
It takes between 3.17 hours and 3.59 hours on my 32bit Windows computers.

They are similar processors within a couple of hundred hertz of each other.

Conan
____________

Message boards : News : New version


Main page · Your account · Message boards


Copyright © 2017 The International Centre for Radio Astronomy Research