Welcome to MilkyWay@home

Are we doing science yet?

Message boards : Number crunching : Are we doing science yet?
Message board moderation

To post messages, you must log in.

AuthorMessage
Jim Wilkins

Send message
Joined: 1 Jan 08
Posts: 40
Credit: 1,676,165
RAC: 0
Message 7711 - Posted: 13 Dec 2008, 15:55:07 UTC

I have lost the bubble here. Are we still in a test phase with the optimized apps or are we doing science with them? Just curious.

Jim
ID: 7711 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 7713 - Posted: 13 Dec 2008, 16:38:11 UTC

I think it is sort of like test-science. Even the tests were supposed to be doing searches if they worked.

The project is supposed to be doing searches, but seems to me the focus isn't on the milkyway part so much as the computer distribution and "fairness".
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 7713 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 7759 - Posted: 15 Dec 2008, 7:42:45 UTC - in response to Message 7713.  

Actually, right now we're comparing the numbers coming in with our asynchronous newton method to what we've done (very slowly) using a synchronous newton method and gradient descent on our cluster. When the current newton method searches (and one more version i'm working on) are working smoothly they should give us more accurate numbers for the current stripes we're computing on (79, 82 and 86) than we have previously had, because what cluster version was using a very inaccurate integral and likelihood calculation.

So yes, in addition to figuring out how to do these accurate local optimization methods asynchronously on BOINC (which is also science), we're doing some real astronomy in getting the most accurate 3-dimensional model of the milkyway galaxy there is (or at least pieces of it).

Once we have the search working correctly, we'll be starting up a whole slew of other strips of the galaxy. I think there's a couple hundred of them that need to be crunched.

After that... there's the problem of putting them all together :D
ID: 7759 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jim Wilkins

Send message
Joined: 1 Jan 08
Posts: 40
Credit: 1,676,165
RAC: 0
Message 7778 - Posted: 15 Dec 2008, 21:06:55 UTC - in response to Message 7759.  

So currently I am seeing a 20x improvement in my crunch times. Is that due to the optimized app, shorter WUs or both?

Thanks,
Jim
ID: 7778 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Emanuel

Send message
Joined: 18 Nov 07
Posts: 280
Credit: 2,442,757
RAC: 0
Message 7779 - Posted: 15 Dec 2008, 23:58:35 UTC

Optimised app. The WUs are different too, but I don't know about shorter.
ID: 7779 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 7780 - Posted: 16 Dec 2008, 0:18:50 UTC

Unless you are talking about the difference from the 10 hour ones to the current?
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 7780 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jim Wilkins

Send message
Joined: 1 Jan 08
Posts: 40
Credit: 1,676,165
RAC: 0
Message 7811 - Posted: 17 Dec 2008, 2:58:02 UTC - in response to Message 7780.  

Yes. Instead of ten hours, I am seeing 30 minutes or less.

Jim
ID: 7811 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile JLDun
Avatar

Send message
Joined: 17 Nov 07
Posts: 77
Credit: 117,183
RAC: 0
Message 7887 - Posted: 20 Dec 2008, 21:43:53 UTC - in response to Message 7778.  
Last modified: 20 Dec 2008, 21:47:31 UTC

So currently I am seeing a 20x improvement in my crunch times. Is that due to the optimized app, shorter WUs or both?

Thanks,
Jim


Keeping in mind that, yes, the old and new apps are processing the workunits differently:

Old App (1.22): ~24h38m (DCF around 48.xxx-49.xxx)
Milksop app (v3): ~0h28m (DCF around 0.93x)

New app (0.7): ~2h50m-3h2m (DCF 0.17x-.19x)

And this is for a two year old Dell Laptop (Intel 1.4GHz Celeron M, MMX SSE SSE2)


[edit]Typo[/edit]
ID: 7887 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Milksop at try

Send message
Joined: 1 Oct 08
Posts: 106
Credit: 24,162,445
RAC: 0
Message 7888 - Posted: 20 Dec 2008, 22:26:07 UTC - in response to Message 7779.  

Optimised app. The WUs are different too, but I don't know about shorter.

The improvement of the new app is larger than what one sees from the shorter crunch times. One should keep in mind that the new WUs are in fact approximately four times as "long" as the old ones. That means they are doing roughly 4 times the work of the old 270 credit WUs.
ID: 7888 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jim Wilkins

Send message
Joined: 1 Jan 08
Posts: 40
Credit: 1,676,165
RAC: 0
Message 7889 - Posted: 20 Dec 2008, 22:33:00 UTC - in response to Message 7888.  

Impressive!
ID: 7889 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Are we doing science yet?

©2024 Astroinformatics Group