A question about RAC


log in

Advanced search

Message boards : Number crunching : A question about RAC

1 · 2 · Next
Author Message
AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 54,669,108
RAC: 22,046
United States
Message 1249382 - Posted: 21 Jun 2012, 12:40:33 UTC

I've read that it takes about 4-6 days for a system's RAC to somewhat stabilize, so I am patiently waiting on my new build http://setiathome.berkeley.edu/show_host_detail.php?hostid=6704589 to respond properly based upon the hardware it has. Does RAC usually rise in a linear fashion? I would have expected 80k or so at least with the current setup, and it's puttering around at about 10-11 now. That would be a huge jump, but then again, maybe it is just a matter of time. It also might be a config issue, using all the latest optimized apps, OC'ing the vid card a tiny bit as well as the CPU (i7-950 - 3.2 from 3.06). Not to be whiny, but sheesh, my Atom with a 295 is beating it right now (14+ k). Thoughts?
____________

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 4665
Credit: 123,795,696
RAC: 95,001
United States
Message 1249388 - Posted: 21 Jun 2012, 12:52:57 UTC
Last modified: 21 Jun 2012, 12:55:20 UTC

RAC generally takes a few weeks to stabilize.

It gets calculated by taking your last average and adding the new credit value.

While not exactly the same as the system that is used. This example may help you.

Day 1 2 3 4 5 6 7 8 9 10 Value 10 25 35 45 45 45 45 45 45 45 Average 10.00 17.50 23.33 28.75 32.00 34.17 35.71 36.88 37.78 38.50

In this example I used the daily numbers on Free-DC for your system and rounded to the nearest 10K. If you continued to produce 45K per day you would see an average of 42k on day 22 and an average of 44k on day 60.
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Profile Fred E.Project donor
Volunteer tester
Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,139,004
RAC: 0
United States
Message 1249398 - Posted: 21 Jun 2012, 13:20:53 UTC

Agree it takes time, but there may be more.

I'm curious about your anon platform APR for MB GPU tasks - 55.0. I've had my 670 installed about 3 weeks and have an APR of 133.9 with a count of .33 and the same driver. It is factory overclocked, but I don't add any more to that. What count are you using? Are you running into temp problems causing you to scale back?

My 3.2 GHZ hexcore is running below your cpu for MB, but mine has been at that level recently and I don't overclock it.
____________
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.

Profile Khangollo
Avatar
Send message
Joined: 1 Aug 00
Posts: 245
Credit: 36,410,524
RAC: 0
Slovenia
Message 1249408 - Posted: 21 Jun 2012, 14:23:17 UTC
Last modified: 21 Jun 2012, 14:35:56 UTC

Now it takes over 2 months for RAC to stabilize on S@H for a new host. Possibly more.
Other projects not that much, but definitely at least one month.

It's not just how it gets calculated over past days but how your completed workunit queue gets validated (completed by your wingmen). Your rac will be more-or-less stable when your "Pending" list won't grow any more and this takes a lot at S@H with 10+10 cache ******* out there and tons of unchecked rogue hosts creating invalid results, causing workunits to linger for months in you "pending" queue.
____________

Profile Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,803,225
RAC: 295
United States
Message 1249429 - Posted: 21 Jun 2012, 15:37:50 UTC
Last modified: 21 Jun 2012, 15:38:14 UTC

+1 to Khangollo, your RAC won't even begin to stabilize until your number of "completed, awaiting validation" stabilizes. When you start growing your RAC, that list of pendings will grow with it, and eventually level out in time.
____________
-Dave #2

3.2.0-33

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1249431 - Posted: 21 Jun 2012, 15:45:26 UTC

If you look on BoincStats you'll see that host has been putting out 30k in the first few days - also likely to go further up as the pendings stabilise.

Now, where do I bugrep that the boincstats link for hosts is broken?
____________
I'm not the Pope. I don't speak Ex Cathedra!

Profile James Sotherden
Avatar
Send message
Joined: 16 May 99
Posts: 9130
Credit: 37,703,074
RAC: 36,441
United States
Message 1249432 - Posted: 21 Jun 2012, 15:45:45 UTC

Back in the day when they used to show total pendings this I7 and its little nvidia GTS 250 had a pending list of anywhere between 50 and 60 K.

The uber crunchers were well over 200k and I believe that over 300K you would get a blank screen. I think Mark had that problem.
____________

Old James

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 4665
Credit: 123,795,696
RAC: 95,001
United States
Message 1249442 - Posted: 21 Jun 2012, 16:09:53 UTC - in response to Message 1249437.

Back in the day when they used to show total pendings this I7 and its little nvidia GTS 250 had a pending list of anywhere between 50 and 60 K.

The uber crunchers were well over 200k and I believe that over 300K you would get a blank screen. I think Mark had that problem.

The pending credits screen has been gone for a long time. Now you can only get a list of your total tasks, and within that you can see the number of pending tasks, but not their potential credit value.
I currently show 7,292 pending tasks, but a lot of the farm has not been doing much crunching due to the heat here, so my pendings and RAC have been declining for the last week or so.

4-6 weeks is the minimum for RAC to begin to stabilize, assuming your crunching and the Seti servers are stable for that length of time.

I was not aware that info by host was broken on Boincstats, might have happened when Willy revamped the site. I generally am looking at the performance of the whole farm, and can see daily credits generated to see what the output was on any given day.

I don't use the machine links very often so it doesn't really bother me.
If the stat guys change the paths I would think they would update the projects to the new location(s). However I think it changed before the site revamp.
The current S@H links point to:
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
With the actual path at the moment being:
http://boincstats.com/en/stats/-1/host/detail/4eac157fa043afe32c2def8e6d8adf56

____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8832
Credit: 53,657,902
RAC: 49,087
United Kingdom
Message 1249451 - Posted: 21 Jun 2012, 16:26:08 UTC - in response to Message 1249442.

Back in the day when they used to show total pendings this I7 and its little nvidia GTS 250 had a pending list of anywhere between 50 and 60 K.

The uber crunchers were well over 200k and I believe that over 300K you would get a blank screen. I think Mark had that problem.

The pending credits screen has been gone for a long time. Now you can only get a list of your total tasks, and within that you can see the number of pending tasks, but not their potential credit value.
I currently show 7,292 pending tasks, but a lot of the farm has not been doing much crunching due to the heat here, so my pendings and RAC have been declining for the last week or so.

4-6 weeks is the minimum for RAC to begin to stabilize, assuming your crunching and the Seti servers are stable for that length of time.

I was not aware that info by host was broken on Boincstats, might have happened when Willy revamped the site. I generally am looking at the performance of the whole farm, and can see daily credits generated to see what the output was on any given day.

I don't use the machine links very often so it doesn't really bother me.
If the stat guys change the paths I would think they would update the projects to the new location(s). However I think it changed before the site revamp.
The current S@H links point to:
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
With the actual path at the moment being:
http://boincstats.com/en/stats/-1/host/detail/4eac157fa043afe32c2def8e6d8adf56

Which of those matches the HostCPID - note, that's not the same thing as a UserCPID - stored in your client_state.xml file? Having looked at one of mine, I suspect that the one being sent by SETI is wrong.

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 4665
Credit: 123,795,696
RAC: 95,001
United States
Message 1249458 - Posted: 21 Jun 2012, 17:00:27 UTC - in response to Message 1249451.
Last modified: 21 Jun 2012, 17:01:11 UTC

Back in the day when they used to show total pendings this I7 and its little nvidia GTS 250 had a pending list of anywhere between 50 and 60 K.

The uber crunchers were well over 200k and I believe that over 300K you would get a blank screen. I think Mark had that problem.

The pending credits screen has been gone for a long time. Now you can only get a list of your total tasks, and within that you can see the number of pending tasks, but not their potential credit value.
I currently show 7,292 pending tasks, but a lot of the farm has not been doing much crunching due to the heat here, so my pendings and RAC have been declining for the last week or so.

4-6 weeks is the minimum for RAC to begin to stabilize, assuming your crunching and the Seti servers are stable for that length of time.

I was not aware that info by host was broken on Boincstats, might have happened when Willy revamped the site. I generally am looking at the performance of the whole farm, and can see daily credits generated to see what the output was on any given day.

I don't use the machine links very often so it doesn't really bother me.
If the stat guys change the paths I would think they would update the projects to the new location(s). However I think it changed before the site revamp.
The current S@H links point to:
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
With the actual path at the moment being:
http://boincstats.com/en/stats/-1/host/detail/4eac157fa043afe32c2def8e6d8adf56

Which of those matches the HostCPID - note, that's not the same thing as a UserCPID - stored in your client_state.xml file? Having looked at one of mine, I suspect that the one being sent by SETI is wrong.

The two ID's were a mistake on my part.
Should have been
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
&
http://boincstats.com/en/stats/-1/host/detail/2054409833ee6d54c92bc069bef54cf4
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8832
Credit: 53,657,902
RAC: 49,087
United Kingdom
Message 1249472 - Posted: 21 Jun 2012, 17:34:44 UTC - in response to Message 1249458.

The two ID's were a mistake on my part.
Should have been
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
&
http://boincstats.com/en/stats/-1/host/detail/2054409833ee6d54c92bc069bef54cf4

Try the link again now - looks like LadyL's bug report has produced a quick fix.

I get (correctly)
http://boincstats.com/en/stats/-1/host/detail/2054409833ee6d54c92bc069bef54cf4
from your host list now.

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 4665
Credit: 123,795,696
RAC: 95,001
United States
Message 1249510 - Posted: 21 Jun 2012, 18:18:36 UTC - in response to Message 1249472.

The two ID's were a mistake on my part.
Should have been
http://boincstats.com/en/stats/-1/detail/2054409833ee6d54c92bc069bef54cf4
&
http://boincstats.com/en/stats/-1/host/detail/2054409833ee6d54c92bc069bef54cf4

Try the link again now - looks like LadyL's bug report has produced a quick fix.

I get (correctly)
http://boincstats.com/en/stats/-1/host/detail/2054409833ee6d54c92bc069bef54cf4
from your host list now.

It looks like they knocked that one out nice and quick. Let's just hope that the location doesn't get changed back anytime soon.

____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8832
Credit: 53,657,902
RAC: 49,087
United Kingdom
Message 1249526 - Posted: 21 Jun 2012, 18:33:35 UTC - in response to Message 1249510.

It looks like they knocked that one out nice and quick. Let's just hope that the location doesn't get changed back anytime soon.

Yes, I've seen Willy change his code to fit a project, and the project change their code to fit Willy, all at the same time....

Took a couple of iterations round that cycle before they agreed on a compromise which worked.

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 54,669,108
RAC: 22,046
United States
Message 1249728 - Posted: 22 Jun 2012, 1:51:29 UTC - in response to Message 1249398.
Last modified: 22 Jun 2012, 1:53:14 UTC

Now it takes over 2 months for RAC to stabilize on S@H for a new host. Possibly more.
Other projects not that much, but definitely at least one month.

It's not just how it gets calculated over past days but how your completed workunit queue gets validated (completed by your wingmen). Your rac will be more-or-less stable when your "Pending" list won't grow any more and this takes a lot at S@H with 10+10 cache ******* out there and tons of unchecked rogue hosts creating invalid results, causing workunits to linger for months in you "pending" queue.

Ok, I can accept the fact that it will take a month or 2 to stabilize to it's relatively final number, but again I would have thought that with 4 of these cards, it would be pushing thru tasks at a much greater rate than it appears to be. My back of the envelope, SWAG would be somewhere around 80ish K? How can one tell if I have some configuration issues going on? Is there something to run or somewhere to look to see if I am running anywhere near where I "should" be for this hardware/software combo?


Agree it takes time, but there may be more.

I'm curious about your anon platform APR for MB GPU tasks - 55.0. I've had my 670 installed about 3 weeks and have an APR of 133.9 with a count of .33 and the same driver. It is factory overclocked, but I don't add any more to that. What count are you using? Are you running into temp problems causing you to scale back?

My 3.2 GHZ hexcore is running below your cpu for MB, but mine has been at that level recently and I don't overclock it.

I am running .25, as these have 2 gig of memory in them, which should allow 4 tasks simultaneously. I am also running 8 tasks on my CPU, for a total of 24 tasks running at one time. I have it slightly OC'ed in Afterburner, +10 on Power Limit, Core Clock and Memory Clock, which seem fairly stable when the temperature is in the 70's like it has been for the last day or so (they are running between 49-56 degrees C). If the temps in the room climb, they sometimes auto down-clock, so I bump them down to +7, or if it's real warm, I set them back to the factory stock settings for the FTW. I think this may have been less challenging if I would have started this in Jan, when it was down around 0 and I could open the window a bit and still stay warm in here... ;-)
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 54,669,108
RAC: 22,046
United States
Message 1249740 - Posted: 22 Jun 2012, 2:29:52 UTC - in response to Message 1249437.

The pending credits screen has been gone for a long time. Now you can only get a list of your total tasks, and within that you can see the number of pending tasks, but not their potential credit value.
Any idea as to why they nuked it? Was it because of the aforementioned blank screen for the uber crunchers, and would have been too much work to fix it, or maybe too much strain on a server?
____________

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1249843 - Posted: 22 Jun 2012, 7:07:15 UTC - in response to Message 1249740.

The pending credits screen has been gone for a long time. Now you can only get a list of your total tasks, and within that you can see the number of pending tasks, but not their potential credit value.
Any idea as to why they nuked it? Was it because of the aforementioned blank screen for the uber crunchers, and would have been too much work to fix it, or maybe too much strain on a server?


No, that was a collateral of the change to 'CreditNew'. Tasks don't have a 'claimed credit' entry any more - credit gets calculated when the tasks validate. If you don't have claimed credit you don't have pending credit.

It may be that you are running too many tasks at once - it's not only a function of memory but also of speeds in other parts of the system. Some people also free a core or two to feed GPUs, especially with multi GPU setups.
You may have to play around to find the optimal values.
____________
I'm not the Pope. I don't speak Ex Cathedra!

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 54,669,108
RAC: 22,046
United States
Message 1250207 - Posted: 23 Jun 2012, 2:12:35 UTC

How do I save a core to do nothing but process GPU requests again? I think I'll try that 1st and see if it makes any difference. Thanks
____________

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,433,321
RAC: 0
Korea, North
Message 1250281 - Posted: 23 Jun 2012, 4:34:46 UTC - in response to Message 1250207.

It doesn't make much of a difference. processing more GPU work will more than make up for any slowdown during the load time for GPU WU's
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

Profile Fred E.Project donor
Volunteer tester
Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,139,004
RAC: 0
United States
Message 1250344 - Posted: 23 Jun 2012, 8:54:35 UTC

How do I save a core to do nothing but process GPU requests again? I think I'll try that 1st and see if it makes any difference. Thanks

In BOINC Mgr's Tools/Computing Preferences/on multiprocessor systems, use at most x % of processors. An alternative is eFMer's Priority tool. I use it to bump GPU tasks to a higher priority than CPU tasks. Perhaps reserving a core or two is worthwhile with that many gpu's.

I have it slightly OC'ed in Afterburner, +10 on Power Limit, Core Clock and Memory Clock, which seem fairly stable when the temperature is in the 70's like it has been for the last day or so (they are running between 49-56 degrees C). If the temps in the room climb, they sometimes auto down-clock, so I bump them down to +7, or if it's real warm, I set them back to the factory stock settings for the FTW. I think this may have been less challenging if I would have started this in Jan, when it was down around 0 and I could open the window a bit and still stay warm in here... ;-)


Are they liquid cooled? Mine runs at 70-76C, and I have to keep the side of the case ajar to keep it that low with ambient temp around 74 F.
____________
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.

1 · 2 · Next

Message boards : Number crunching : A question about RAC

Copyright © 2014 University of California