Thread 'Question about CPU Usage percentage'

Message boards : Questions and problems : Question about CPU Usage percentage
Message board moderation

To post messages, you must log in.

AuthorMessage
droople
Avatar

Send message
Joined: 20 Aug 08
Posts: 26
China
Message 19673 - Posted: 22 Aug 2008, 2:17:56 UTC

Hi

I read the FAQ from here http://boincfaq.mundayweb.com/index.php?language=1&view=45&sessionID=c990e540161dd54c6be7f8a7e35e320a

It says "Because BOINC does not use the Windows registry,plus because it is a multi platform program, it cannot use Windows specific timing APIs as other 3rd party programs such as Threadmaster use."

Does this mean boinc.exe or individual project application can't use the APIs?

I'm also using Folding@home, which is also multi platform, and it seems that it can use the APIs.

Cheers
ID: 19673 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19675 - Posted: 22 Aug 2008, 2:30:20 UTC - in response to Message 19673.  

It means there is only one source code that for all platforms to download and compile from. When you then compile it for Windows, Linux or OS-X (or any other platform), you can't have dependencies on an API that's only in use by one of the platforms.

The example I gave, Threadmaster, uses a key in the Windows registry to set a maximum amount of CPU cycles for the application to use. If BOINC were to be dependent on this same key, it couldn't be built from the same source code for other OSes that don't use a registry.

ID: 19675 · Report as offensive
droople
Avatar

Send message
Joined: 20 Aug 08
Posts: 26
China
Message 19678 - Posted: 22 Aug 2008, 3:48:31 UTC - in response to Message 19675.  

Hi Ageless

Thank you for your explanation.

Dose this mean that BOINC developers think it is unnecessary or no benefit to modify source code for each platform?

I'm new to distribution computing, actually I installed BOINC and folding@home on my laptop this Wednesday.

The WU of folding@home will be finished today, the total CPU time is about 260 mins, and the average CPU usage percentage is 8% (Shown by Process Lasso).

I put the BOINC CPU Usage percentage to 5%, and Process Lasso shows the average CPU usage is around 4%, the total CPU time is about 90 mins. Every time the application run, I can hear my CPU fan speed increase and then decrease.

For my personal understanding, folding@home uses CPU more efficiently than BOINC. And I'm reluctant to increase the CPU Usage for BOINC, because I don't want my laptop become a hair dryer.

Cheers

ID: 19678 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19681 - Posted: 22 Aug 2008, 12:17:50 UTC - in response to Message 19678.  

Dose this mean that BOINC developers think it is unnecessary or no benefit to modify source code for each platform?

They are modifying the source code so it works on each platform. What they aren't doing is give one source code for Windows, one for Linux, one for OS-X and one for all the rest. It's all just one source code, but with different build options.

I put the BOINC CPU Usage percentage to 5%, and Process Lasso shows the average CPU usage is around 4%, the total CPU time is about 90 mins. Every time the application run, I can hear my CPU fan speed increase and then decrease.

First off, it's very unwise to install and Folding and BOINC on one system as both will try to get the left-over CPU cycles. Don't run them side-by-side, you will get problems.

Second, setting BOINC to use 5% CPU means that on a daily basis, you allow it to run about an hour. Seeing that most projects have very strict deadlines by which the task should be in, you're not going to be able to get much work done. For instance, some projects have tasks that run 8 to 12 hours, yet have a 3 day deadline. By day 3, you have run 3 hours and are already outside the deadline.

So don't do this. The percentage you set is also not continuous 5%, it means that BOINC pauses the application for so many seconds and lets it run for so many seconds. On a "normal" 50%, this means that per minute, 30 seconds will be run at full speed (100% CPU) and 30 seconds will be idle (0% CPU). Your 5% means it'll run 3 seconds and stay 57 seconds idle. Not very efficient!

If you don't want your fan to sound like a fighter jet thundering down the run-way, either don't run BOINC on the laptop, or use the third party throttler Threadmaster.

Do know that because of what they do, both FAH and BOINC's projects will put your CPU under stress, causing it to heat up more than it would by opening Word or other normal programs.
ID: 19681 · Report as offensive
droople
Avatar

Send message
Joined: 20 Aug 08
Posts: 26
China
Message 19684 - Posted: 22 Aug 2008, 13:46:53 UTC - in response to Message 19681.  

Hi Ageless

Thank you very much for your advice.

I'm curious about the projects you mentioned that have tasks that run 8 to 12 hours, yet have a 3 day deadline. Could you give me some examples, I will avoid these projects.

Cheers
ID: 19684 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19685 - Posted: 22 Aug 2008, 14:45:05 UTC - in response to Message 19684.  

Milkyway, Malaria, some of the really short ones at Seti run for a couple of hours on a 7 day deadline.
ID: 19685 · Report as offensive
Ricky Moore

Send message
Joined: 26 Dec 06
Posts: 36
United States
Message 19697 - Posted: 23 Aug 2008, 0:55:07 UTC

I have a question about this CPU % useage.... It seems to me that BOINC does not take into account the fact that it only uses the CPU at like 80%.

I have a Dell Insp 2200 that overheats at 100% useage so I lowered it to 75%. However BOINC still downloads the same number of WU's as it did when running at 100% of the CPU useage. This sometimes causes problems because the WU's do bot get done by the deadline.

ID: 19697 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1179
Argentina
Message 19699 - Posted: 23 Aug 2008, 1:59:16 UTC - in response to Message 19681.  

Dose this mean that BOINC developers think it is unnecessary or no benefit to modify source code for each platform?

They are modifying the source code so it works on each platform. What they aren't doing is give one source code for Windows, one for Linux, one for OS-X and one for all the rest. It's all just one source code, but with different build options.

But this is just one feature. It's not hard to:
#ifdef WIN32
/* use the Windows-specific way to throttle */
#else
/* current 1-second-interval throttling */
#endif

In fact that's already done in dozens of places. Even seemingly-simple things like copying a file has different code for Windows.
ID: 19699 · Report as offensive
droople
Avatar

Send message
Joined: 20 Aug 08
Posts: 26
China
Message 19701 - Posted: 23 Aug 2008, 2:55:18 UTC - in response to Message 19685.  

Milkyway, Malaria, some of the really short ones at Seti run for a couple of hours on a 7 day deadline.


uhh, sounds interesting.
ID: 19701 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19702 - Posted: 23 Aug 2008, 5:53:05 UTC - in response to Message 19697.  

However BOINC still downloads the same number of WU's as it did when running at 100% of the CPU useage.

The CPU Usage function is a thermal throttle. As far as I know it has no integration into the work-fetch policy module of BOINC. Perhaps that eventually, after long enough use, that you'll see work-fetch decrease, but not from minute one.

If you don't want as much work as you're getting, please decrease your additional work buffer.
ID: 19702 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19703 - Posted: 23 Aug 2008, 5:55:46 UTC - in response to Message 19699.  

But this is just one feature.

I'm not expecting the devs (or you with your fork) to go figure out and add different ways of thermal throttling BOINC (or synecdoche) in the various forms of Linux that are around, all with their own way of doing things. The amount of if-endif's would be infinite... and that for one little module only. ;-)
ID: 19703 · Report as offensive
Keck_Komputers
Avatar

Send message
Joined: 29 Aug 05
Posts: 304
United States
Message 19704 - Posted: 23 Aug 2008, 6:49:21 UTC - in response to Message 19702.  

The CPU Usage function is a thermal throttle. As far as I know it has no integration into the work-fetch policy module of BOINC. Perhaps that eventually, after long enough use, that you'll see work-fetch decrease, but not from minute one.

It takes about a month for the work-fetch policy to adjust for throttling. It should make your % that BOINC is allowed to work drop down to approximately match the throttle setting. That will reduce the work fetch.
BOINC WIKI

BOINCing since 2002/12/8
ID: 19704 · Report as offensive
Ricky Moore

Send message
Joined: 26 Dec 06
Posts: 36
United States
Message 19710 - Posted: 23 Aug 2008, 10:41:53 UTC - in response to Message 19708.  

The CPU Usage function is a thermal throttle. As far as I know it has no integration into the work-fetch policy module of BOINC. Perhaps that eventually, after long enough use, that you'll see work-fetch decrease, but not from minute one.

It takes about a month for the work-fetch policy to adjust for throttling. It should make your % that BOINC is allowed to work drop down to approximately match the throttle setting. That will reduce the work fetch.

Is this advice based on latest clients or from longer past? I use Threadmaster GUI at 70% this moment on a quad and the DCF hiked to 1.44 rather rapidly and is staying in that area (Coincidence?). Maybe quads learn quicker than single core P4s ?!?


I find it strange that a project like.... Milkyway can tell you that a WU won't finish in time because it "only gets 50%" of the computer time but it takes BOINC a month to learn it has to download less work in order to meet the deadlines. Meanwhile it download WU's it can't finish in time and you don't always know until it's to late to do anything. I lost almost 30 WU's because of this.

I thank you for letting me know about this. Maybe in upcoming changes somthing can be written into BOINC to adjust for that. Meanwhile I'll reduce the number of days down.



ID: 19710 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15539
Netherlands
Message 19711 - Posted: 23 Aug 2008, 11:05:26 UTC - in response to Message 19710.  

I find it strange that a project like.... Milkyway can tell you that a WU won't finish in time because it "only gets 50%" of the computer time but it takes BOINC a month to learn it has to download less work in order to meet the deadlines.

First off, Milkyway only just (6 days ago) changed their deadline from 2 days up to 3 days. If you want to teach BOINC things like this, take into account at least 14 days.

Second, that message that you see should tell you that you won't get work at all.
It's a reply of the scheduler telling you that the amount of work you are asking for can impossibly be done within the deadline due to the limitations on your computer.

Message from server: No work sent (won't finish in time) Computer on 'x'% of time, BOINC on 'y'% of that, this project gets 'z'% of that is the complete message.
I'd like to see your line against it...
ID: 19711 · Report as offensive
Keck_Komputers
Avatar

Send message
Joined: 29 Aug 05
Posts: 304
United States
Message 19720 - Posted: 23 Aug 2008, 20:26:51 UTC - in response to Message 19708.  

The CPU Usage function is a thermal throttle. As far as I know it has no integration into the work-fetch policy module of BOINC. Perhaps that eventually, after long enough use, that you'll see work-fetch decrease, but not from minute one.

It takes about a month for the work-fetch policy to adjust for throttling. It should make your % that BOINC is allowed to work drop down to approximately match the throttle setting. That will reduce the work fetch.

Is this advice based on latest clients or from longer past? I use Threadmaster GUI at 70% this moment on a quad and the DCF hiked to 1.44 rather rapidly and is staying in that area (Coincidence?). Maybe quads learn quicker than single core P4s ?!?

Based on version 5 clients, however I don't think the time stats code was changed for version 6. Threadmaster will cause a different stat to change; cpu effeciency. The end result is the same though.

The DCF is a whole other calculation. It goes up quickly but down slowly and it is easier to estimate how long it will take as a number of tasks. This one reflects how accurately the project has estimated the difficulty of their tasks.
BOINC WIKI

BOINCing since 2002/12/8
ID: 19720 · Report as offensive

Message boards : Questions and problems : Question about CPU Usage percentage

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.