Longer target CPU settings??

Message boards : Number crunching : Longer target CPU settings??

To post messages, you must log in.

AuthorMessage
Profile paulcsteiner

Send message
Joined: 15 Oct 05
Posts: 19
Credit: 3,120,062
RAC: 968
Message 42128 - Posted: 13 Jun 2007, 3:02:46 UTC

So I,m up very late (or early as the case may be) drinking coffee and eating chocolate, surfing the Forum and I start thinking what would happen if the setting for target CPU usage increased from a max of 24 hours to maybe 48 or longer?
Would this be of any use to Dr. Baker & Co.? Would the longer run time produce better results,or perhaps a higher resolution on the folding predictions? Would there be anybody that would want to have a work unit with a two day(or longer) work unit?
ID: 42128 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
MattDavis
Avatar

Send message
Joined: 22 Sep 05
Posts: 206
Credit: 1,377,748
RAC: 0
Message 42129 - Posted: 13 Jun 2007, 3:09:36 UTC
Last modified: 13 Jun 2007, 3:10:12 UTC

A 24 hour unit would do the same amount of science as 6 4 hour units, or 2 12 hour units, etc. etc.

The difference:

You would be doing all the science in one kind of unit with a 24 hour unit, while you will probably get different kinds of units if you did 6 4 hour units.

Also, you'll reduce server load on the Rosetta servers and decrease your own bandwidth usage with bigger units.
ID: 42129 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile paulcsteiner

Send message
Joined: 15 Oct 05
Posts: 19
Credit: 3,120,062
RAC: 968
Message 42130 - Posted: 13 Jun 2007, 3:33:24 UTC

Hi MattDavis,
Right, so the science in a theoretical 48 hour WU would be greater? or just more?(either way seems good)
I'm set at 24 for all my machines, including everything from some really "old school" hardware, to a fairly sweet dual-dual core Xeon. I noticed that the amount of decoys generated by my machines vary significantly from machine to machine in relation to the computing potential of each machine. For the high end machines, I wonder if a 24 hour WU is limiting?
ID: 42130 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
MattDavis
Avatar

Send message
Joined: 22 Sep 05
Posts: 206
Credit: 1,377,748
RAC: 0
Message 42131 - Posted: 13 Jun 2007, 3:46:30 UTC - in response to Message 42130.  

Hi MattDavis,
Right, so the science in a theoretical 48 hour WU would be greater? or just more?(either way seems good)
I'm set at 24 for all my machines, including everything from some really "old school" hardware, to a fairly sweet dual-dual core Xeon. I noticed that the amount of decoys generated by my machines vary significantly from machine to machine in relation to the computing potential of each machine. For the high end machines, I wonder if a 24 hour WU is limiting?


The science in a 48 hour WU is the exact same as 2 24 hour units, or 4 12 hour units, or 8 6 hour units, etc. etc.

A longer runtime doesn't do any more science than several smaller runtimes that equal the same longer runtime.
ID: 42131 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
hugothehermit

Send message
Joined: 26 Sep 05
Posts: 238
Credit: 314,893
RAC: 0
Message 42132 - Posted: 13 Jun 2007, 6:35:59 UTC
Last modified: 13 Jun 2007, 6:53:54 UTC

G'day paulcsteiner

here

There used to be a 4 day limit, but they changed it to 1 day as they were getting the 1% stuck bug, which was causing a lot of trouble.

As for the difference between the amount of hours per WU and science done, minuscule if anything at all, it would only help with bandwidth.

More WU's less time or more time less WU's it adds up to the same amount of models/decoys (science) done*.

Even if they implemented an algorithm that took your last models/decoys into account in the search space, it's such a big search space, one computer can't effectively do enough to make any meaningful difference (I'm guessing).

Edit: spelling, and a bit more spelling, and to try to make it readable.
Edit: *this isn't exactly accurate, as different WU's have different calculation times per model/decoy but it's the best I could do, without going into the different credits for different WU's etc... I hope it helps anyway.
ID: 42132 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile paulcsteiner

Send message
Joined: 15 Oct 05
Posts: 19
Credit: 3,120,062
RAC: 968
Message 42160 - Posted: 14 Jun 2007, 0:16:22 UTC - in response to Message 42132.  

Hi Hugo,

Thanks very much for the info. I'm starting to wrap my head around the concept. It just seemed to me that longer run times should generate more decoys, more decoys, the better?. I dunno, I gotta do some reading ands catch up to you's guys in the know!, Cheers and happy crunching.


G'day paulcsteiner

here

There used to be a 4 day limit, but they changed it to 1 day as they were getting the 1% stuck bug, which was causing a lot of trouble.

As for the difference between the amount of hours per WU and science done, minuscule if anything at all, it would only help with bandwidth.

More WU's less time or more time less WU's it adds up to the same amount of models/decoys (science) done*.

Even if they implemented an algorithm that took your last models/decoys into account in the search space, it's such a big search space, one computer can't effectively do enough to make any meaningful difference (I'm guessing).

Edit: spelling, and a bit more spelling, and to try to make it readable.
Edit: *this isn't exactly accurate, as different WU's have different calculation times per model/decoy but it's the best I could do, without going into the different credits for different WU's etc... I hope it helps anyway.


ID: 42160 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
AlphaLaser

Send message
Joined: 19 Aug 06
Posts: 52
Credit: 3,327,939
RAC: 0
Message 42164 - Posted: 14 Jun 2007, 5:16:39 UTC

Unless you want to be really picky and count the second or less it takes BOINC to end one result and start the next ;)
ID: 42164 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
MattDavis
Avatar

Send message
Joined: 22 Sep 05
Posts: 206
Credit: 1,377,748
RAC: 0
Message 42192 - Posted: 15 Jun 2007, 1:04:23 UTC

I've tried to explain it several times but I don't see a more simple way to do it.
ID: 42192 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Feet1st
Avatar

Send message
Joined: 30 Dec 05
Posts: 1755
Credit: 4,690,520
RAC: 0
Message 42193 - Posted: 15 Jun 2007, 2:44:44 UTC

How about this:

4 hrs... 24 models;
12 hrs... 72 models;

... making predictor of the day,
... priceless!

You see Rosetta needs 10,000-100,000 models crunched for most proteins and methods of study in order to have meaningful results. If you crunch 3 tasks in 12 hours, or 1, you are still going to have about 72 models (all other things being equal as to type of protein you are crunching etc.).
Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/
ID: 42193 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile paulcsteiner

Send message
Joined: 15 Oct 05
Posts: 19
Credit: 3,120,062
RAC: 968
Message 42218 - Posted: 16 Jun 2007, 1:06:35 UTC - in response to Message 42193.  

HAHAAA~!
Yes perfect. I keep hoping to make POTD.


How about this:

4 hrs... 24 models;
12 hrs... 72 models;

... making predictor of the day,
... priceless!

You see Rosetta needs 10,000-100,000 models crunched for most proteins and methods of study in order to have meaningful results. If you crunch 3 tasks in 12 hours, or 1, you are still going to have about 72 models (all other things being equal as to type of protein you are crunching etc.).


ID: 42218 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : Longer target CPU settings??



©2024 University of Washington
https://www.bakerlab.org