Advanced search

Message boards : Graphics cards (GPUs) : Why is there a work unit quota?

Author Message
JZ
Send message
Joined: 1 Sep 08
Posts: 1
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 2199 - Posted: 8 Sep 2008 | 19:17:46 UTC

I had the system configured wrong so I received computational errors when attempting to crunch. I believe I have it set up correctly now, but the client won't give me any more work units because it says I have already reached my daily quota of 4 (quad core system). What is the rationale for limiting how much work people can do in a day? Will this limit be waived in the future? Why not let people crunch as much as they want?

Profile Venturini Dario[VENETO]
Send message
Joined: 26 Jul 08
Posts: 44
Credit: 4,832,360
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwat
Message 2201 - Posted: 8 Sep 2008 | 19:26:11 UTC - in response to Message 2199.

I had the system configured wrong so I received computational errors when attempting to crunch. I believe I have it set up correctly now, but the client won't give me any more work units because it says I have already reached my daily quota of 4 (quad core system). What is the rationale for limiting how much work people can do in a day? Will this limit be waived in the future? Why not let people crunch as much as they want?


I suppose it is because the project doesn't produce many workunits so they try to "spread" them as much as possible

Profile Stefan Ledwina
Avatar
Send message
Joined: 16 Jul 07
Posts: 464
Credit: 135,911,881
RAC: 892
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 2203 - Posted: 8 Sep 2008 | 19:37:52 UTC - in response to Message 2199.

...What is the rationale for limiting how much work people can do in a day? ...


Because if there wouldn't be a limit people with misconfigured computers could produce hundrets or thousands computeational errors per day, which would slow down the whole projects because of resends.
AFAIK every BOINC projects has a limit set. Some have a higher limit therefore you don't really see it, others have a lower limit...
____________

pixelicious.at - my little photoblog

Profile Deleted
Avatar
Send message
Joined: 7 Sep 08
Posts: 6
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 2244 - Posted: 11 Sep 2008 | 2:03:21 UTC - in response to Message 2203.

> if there wouldn't be a limit people with misconfigured computers could produce hundrets or thousands computeational errors per day, which would slow down the whole projects because of resends.

Uhhh... 'scuse me! Shouldn't the quota be set HIGHER than any user could legitimately process? Has it occurred to "management" that they're throwing away precious resources?

Has it occurred to them to have the software notice when a user is error flooding, lock out just that user, and send the user an email about it?

128 shader units blast through my quota in 6 hours. If my GPU is going to sit idle 18 hours a day, then I'm gonna go back to rosetta@home on my main CPU.

Who can I email about this who has the power to reevaluate this policy?

--flk

____________
A cerebral mutant, damned by God and doomed by man to understand EVERYTHING and control NOTHING.

My blog

Profile koschi
Avatar
Send message
Joined: 14 Aug 08
Posts: 124
Credit: 466,579,198
RAC: 297,582
Level
Gln
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 2246 - Posted: 11 Sep 2008 | 2:45:51 UTC
Last modified: 11 Sep 2008 | 2:46:49 UTC

The quota only applies if one is returning back to much error units.
If you would build a card that computes a unit in 1 hours and validates without problems, I assume it would be no problem for the project to hand out at least 24 units each day. But if your computer errors out each unit within seconds, then the system will not give you more work. what for?

I'm not speaking about 1 or 2 units which can always fail, due to various reasons. Today I had to kill 2 units that where close to the deadline, but I got new ones. If I would have killed them also, maybe I wouldn't get more for that day.

This is a simple technique to protect the project. As each work unit is ~8MB right now, imagine some dozens of users (that might not even know) whose machines are aborting each and every work unit and hence loading more and more units. This could easily degrade the webservers performance. Additionally every unit given to some user creates entries in die projects database. An extreme was the milkyway@home project. They had units that completed within few minutes, so one computer could calculate several hundreds of units each day. Take this and multiply it by some thousand users and the database would grow like insane...
To stand this, they gave only 8 units to each computer at the same time and deleted old DB entries within 5 minutes after the WU was validated.

The last point is not the problem on this project, but just one example what BOINC admins have to take care of.
It's not that want to frustrate users, that would be stupid, as they would disappear quickly...

So if you manage to solve the problem your computer has with this project, and it will finally return valid results, you'll get enough work to process ;-)

Profile Stefan Ledwina
Avatar
Send message
Joined: 16 Jul 07
Posts: 464
Credit: 135,911,881
RAC: 892
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 2248 - Posted: 11 Sep 2008 | 4:37:00 UTC - in response to Message 2244.



...
128 shader units blast through my quota in 6 hours. If my GPU is going to sit idle 18 hours a day, then I'm gonna go back to rosetta@home on my main CPU.
...
--flk


Maybe because every singly WU on your computer errored out!? I saw in the stderr.out of your WUs that you use a 8800GTS, is that right?
If so, your card is not compatible to the current app. Please have a look at this Thread in the FAQ to know why.
____________

pixelicious.at - my little photoblog

Profile koschi
Avatar
Send message
Joined: 14 Aug 08
Posts: 124
Credit: 466,579,198
RAC: 297,582
Level
Gln
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 2250 - Posted: 11 Sep 2008 | 4:48:44 UTC
Last modified: 11 Sep 2008 | 4:50:59 UTC

I don't think that this is the problem here, as there are three indicators of a compatible G92 chip :)

- 128 shaders
- shader clock speed of 1620MHz which is 400MHz above G80 based GTS level
- Device 0: "GeForce 8800 GTS 512" string and all 512MB GTS cards are based on G92

Maybe some driver issue, not Cuda 2.0 capable or like that...

Profile [SETI.USA]Tank_Master
Avatar
Send message
Joined: 8 Jul 07
Posts: 85
Credit: 67,463,387
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 2252 - Posted: 11 Sep 2008 | 5:23:12 UTC

there really isn't need for more than the current amount of work, for most people here who can run this project have 1 graphics card and a dual or quad core computer. They then have 2 or 4 WUs, but can only run one at at time, so there is plenty of time to download another WU before running out.

This will all be a mute point when this project goes into non_CPU_Intensive mode, for there can be a set number of WUs per host of say 7 or whatever.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 2253 - Posted: 11 Sep 2008 | 7:45:13 UTC

There are 2 kinds of quotas. On is the limit for the amount of WUs one is allowed to cache. That's the limit set by the number of CPUs and the current minimum amount of work / maximum amount of GPUs you could possibly use.

But as soon as you return one WU you're allowed to get another one. The overall daily limit can be found in the profile, I think the current maximum is 8. For now that's sufficient even for dual 280GTX. If people start using 3 of them or 3 9800GX2 this quota could easily be increased.

MrS
____________
Scanning for our furry friends since Jan 2002

Thamir Ghaslan
Send message
Joined: 26 Aug 08
Posts: 55
Credit: 1,475,857
RAC: 0
Level
Ala
Scientific publications
watwatwat
Message 2262 - Posted: 11 Sep 2008 | 15:43:12 UTC - in response to Message 2250.

I don't think that this is the problem here, as there are three indicators of a compatible G92 chip :)

- 128 shaders
- shader clock speed of 1620MHz which is 400MHz above G80 based GTS level
- Device 0: "GeForce 8800 GTS 512" string and all 512MB GTS cards are based on G92

Maybe some driver issue, not Cuda 2.0 capable or like that...


The project "banned" me for a day after I error. 4 task rule in action.

I increased the core clock to from 602 to 704 MHZ and the memory from 1107 to 1258, giving a shader from 1296 to 1516.

It computed to almost 70% without any problems and then it went whack!

At 100% fan speed the core was at 63C.

I'd be interested to know any success stories from any GTX280 overclocks in here! Otherwise, stock speed would suffice!

Profile koschi
Avatar
Send message
Joined: 14 Aug 08
Posts: 124
Credit: 466,579,198
RAC: 297,582
Level
Gln
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 2266 - Posted: 11 Sep 2008 | 16:56:51 UTC

Indiana74 from our team has an overclocked GTX280, running at 684Mhz chip / 1458Mhz shader / 1296Mhz memory. Times are around 25100 seconds...

Profile Deleted
Avatar
Send message
Joined: 7 Sep 08
Posts: 6
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 2273 - Posted: 11 Sep 2008 | 19:01:37 UTC - in response to Message 2266.


Okay, I finally located the problem. It wasn't my G92 chip (the card was O/C'd at the factory and is stable); it was a spurious error in one of my memory modules that has been causing my system to screw up and sometimes crash for a couple of weeks. It is a one-bit error which occurs only now-and-then, so it's hard to tell when an app has been corrupted (apparently BOINC detects it every time it happens, which is great).

I haven't replaced that mem yet, but when I do, how will I get a normal quota again? Do I have to request it, or what?

thanx,

flk


____________
A cerebral mutant, damned by God and doomed by man to understand EVERYTHING and control NOTHING.

My blog

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 2274 - Posted: 11 Sep 2008 | 19:01:53 UTC
Last modified: 11 Sep 2008 | 19:04:28 UTC

@Koschi: Indiana74 also gets some calculation errors.. not sure if I'd call this OC a success.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Stefan Ledwina
Avatar
Send message
Joined: 16 Jul 07
Posts: 464
Credit: 135,911,881
RAC: 892
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 2280 - Posted: 11 Sep 2008 | 19:27:32 UTC - in response to Message 2273.
Last modified: 11 Sep 2008 | 19:28:05 UTC


....

I haven't replaced that mem yet, but when I do, how will I get a normal quota again? Do I have to request it, or what?

thanx,

flk




No, you don't have to request it. With every successful task you return to the project server, your daily quota will go up again...
____________

pixelicious.at - my little photoblog

Profile Krunchin-Keith [USA]
Avatar
Send message
Joined: 17 May 07
Posts: 512
Credit: 111,288,061
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 2285 - Posted: 11 Sep 2008 | 23:45:30 UTC - in response to Message 2199.

I had the system configured wrong so I received computational errors when attempting to crunch. I believe I have it set up correctly now, but the client won't give me any more work units because it says I have already reached my daily quota of 4 (quad core system). What is the rationale for limiting how much work people can do in a day? Will this limit be waived in the future? Why not let people crunch as much as they want?


The quota is adjusted automatically by the boinc server.

The quota is to protect the project from hosts only returning errors. It limits the erred units until they are fixed so that these erred units do not quickly consume all the work needed by functional hosts. Too many errors returned on a work unit will cause it to not be processed any more. This causes problems for the project as it has to then reissue the work again. Too many hosts with errors could quickly consume all the work available, mark it all invalid and no-one would have work, thus slowing down the entire process making it a bad experience for everyone.

The quota starts at a fixed number per cpu per day.

If your host returns error, your quota is reduced.

When your host begins to return valid results, your quota is raised.

There are limits each project can set.

To find your quota, go to the computers list under your account. click on one of the host id numbers. You will see somewhere on that page a line similar to this:
Maximum daily WU quota per CPU 8/day

The reason it is based on CPUs' is because before this project, there were only CPU's on BOINC. This is the way the BOINC developers (not the project) designed the system.

Profile Edboard
Avatar
Send message
Joined: 24 Sep 08
Posts: 72
Credit: 12,410,275
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 2864 - Posted: 7 Oct 2008 | 21:53:35 UTC

I have just installed 2 Nvidia GTX280 in my PC running under Vista 32 bits and, after doing all needed stuff (extend desktop, etc) I have gotten it to work properly. But I have a problem:

My CPU is a Core 2 Duo, and I only have two workunits at anytime, as I had when I had only one GPU. The problem is that when one WU is done, the GPU has to wait until a new one is downloaded because the other one is being crunched by the other GPU. The last time, one GPU was 45 minutes doing nothing, because there was any WU to do.

NOTE: I have NO error unit in this PC that can limit me to a low quota.

Profile Kokomiko
Avatar
Send message
Joined: 18 Jul 08
Posts: 190
Credit: 24,093,690
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 2866 - Posted: 7 Oct 2008 | 22:09:54 UTC - in response to Message 2864.

The number of WUs is limited by the number of cores of your CPU. If you own a Quad-core, you get 4 WUs at the time. At the moment the CUDA-supportet BOINC-version is experimental and in beta test. The only way top get more is at the moment to change the CPU to a Quad or Triple-core CPU.
____________

Profile Edboard
Avatar
Send message
Joined: 24 Sep 08
Posts: 72
Credit: 12,410,275
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 2868 - Posted: 7 Oct 2008 | 22:48:13 UTC - in response to Message 2866.
Last modified: 7 Oct 2008 | 22:49:12 UTC

Well, I have no intention to change my recently buyed Core 2 Duo, so I'll wait until BOINC allows to have more WUs downloaded at a time. In the meanwhile, it would be interesting for me to have only one GPU in GPUGRID, so I can have the other one crunching for FOLDING. My question is:

Can I limit GPUGRID to work only in one GPU in a two GPU PC?

If it is possible: How can I do it?

Post to thread

Message boards : Graphics cards (GPUs) : Why is there a work unit quota?

//