Welcome to MilkyWay@home

Configuring BOINC

Message boards : Number crunching : Configuring BOINC
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Rick Painter

Send message
Joined: 2 Mar 09
Posts: 3
Credit: 8,544,464
RAC: 0
Message 25215 - Posted: 13 Jun 2009, 0:57:10 UTC

Using BOINC 6.6.31.

I'm trying to get SETI and MW to run together but SETI keeps trying to hog all the resources. I would like 3 cores running SETI and one supporting the graphics card running MW. Can anyone recommend configs?

I seem to remember that this was easier with earlier versions of BOINC.
ID: 25215 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile The Gas Giant
Avatar

Send message
Joined: 24 Dec 07
Posts: 1947
Credit: 240,884,648
RAC: 0
Message 25244 - Posted: 13 Jun 2009, 11:49:45 UTC

For my quad I try to keep 4 tasks running on my cpu and 4 tasks running across me 2 gpu's. To achieve this I give MW a resource share a smidge under 50% (currently 49.66%) and then split the rest of the resource share across the projects I want to run. I have down graded BOINC to 6.4.5 as I think the work fetch and work scheduling does not work appropriately with the post 6.6.20 versions - but then I haven't tried the latest ones with my current set up.
ID: 25244 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MontagsMeeting

Send message
Joined: 12 Mar 09
Posts: 61
Credit: 9,214,340
RAC: 0
Message 25252 - Posted: 13 Jun 2009, 13:12:12 UTC

I try to figure this out for a long time now and with the availability of WUs the situation becomes worse.
I have a core i7 that should run Seti on all cores and a 4850 that should run MW, so far so good my core i7 is running 8 Seti-tasks to achieve this i have to set MW to <avg_ncpus>0.01</avg_ncpus> in app_info.xml. With this avg_ncpus setting every downloaded MW WU will start immediately. On the gpu only 2 of all running MW WUs are really calculated the other >50 tasks stay in memory - with about 15MB per task this makes 700-800MB.
Problem is, that things aren't very stable this way and from time to time MW stops 1 or 2 tasks so i have a lot of active tasks in memory waiting for nothing to happen.

I've tried every setting i'm aware of but nothing heales this behavior
ressource share does absolutely nothing for me
avg_ncpus is the real share-setting anything higher than 0.01 will drop one cpu core after the other
cache-settings too do nothing
ID: 25252 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 25270 - Posted: 13 Jun 2009, 16:30:31 UTC
Last modified: 13 Jun 2009, 16:40:50 UTC

I have only been using a GPU for a few days. Mine does the same. I have a Xeon W3520 (=i7) and a 3850. All 48 WUs stay in memory. What causes errors is that they are not processed in order. This means if I let it go by itself overnight many of the WUs time out. The newer double wedge thickness WUs make it worse because they take 170 seconds each instead of 85, but the single wedge thickness WUs that are part of the cache of 48 still time out after an hour.

I am very pleased with the speed of processing compared to a CPU so I'm not complaining, just pointing out that these errors are inefficient and waste bandwidth because the timed out WUs have to be sent out again to someone else. I understand this would not happen as much to those with faster/multiple graphics cards and/or computers with 4 cores or less. I could buy a few 4770s but they will not work with XP drivers. Better Stream support from ATI for their newer graphics cards would be helpful. I suppose only an application integrated into BOINC will enable WUs to be processed in the order in which they are received.

I was getting a few freezes when processing 3 or 2 at a time, but now with <cmdline>n1</cmdline> it is very stable. The application itself works very well for me only the timeouts give trouble.
ID: 25270 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 25272 - Posted: 13 Jun 2009, 16:48:44 UTC - in response to Message 25252.  

I try to figure this out for a long time now and with the availability of WUs the situation becomes worse.
I have a core i7 that should run Seti on all cores and a 4850 that should run MW, so far so good my core i7 is running 8 Seti-tasks to achieve this i have to set MW to <avg_ncpus>0.01</avg_ncpus> in app_info.xml. With this avg_ncpus setting every downloaded MW WU will start immediately. On the gpu only 2 of all running MW WUs are really calculated the other >50 tasks stay in memory - with about 15MB per task this makes 700-800MB.
Problem is, that things aren't very stable this way and from time to time MW stops 1 or 2 tasks so i have a lot of active tasks in memory waiting for nothing to happen.

I've tried every setting i'm aware of but nothing heales this behavior
ressource share does absolutely nothing for me
avg_ncpus is the real share-setting anything higher than 0.01 will drop one cpu core after the other
cache-settings too do nothing

You should really set the avg_ncpus value higher. That way a lot less WUs are loaded and also the time-out issue is no longer present. If you loose one or two tasks is not much of an issue with a Core i7 because of the hyperthreading. And if you really want to avoid it use the cc_config.xml to raise the number of CPUs the BOINC client sees. If you set it to 9 and adjust your MW resource share to 11.1% (1/9) you can more or less reserve one of the 9 CPUs for MW and can afford a lot higher avg_ncpus values (you can even set it to 1.0). At least if you reset the debts it works quite well here.

But generally speaking the scheduler of the BOINC client is quite a mess in my opinion. The features of the newer clients (6.x.xx) like fractional CPU allocation are simply not working as they should.
ID: 25272 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MontagsMeeting

Send message
Joined: 12 Mar 09
Posts: 61
Credit: 9,214,340
RAC: 0
Message 25296 - Posted: 13 Jun 2009, 19:47:03 UTC

Today i tried different settings with a ressource share down to 1%, avg_ncpu up to 1 and up to 10 cpus but only a avg_ncpu setting of 0.01 gives me full cpu usage and as Seti has higher priority than MW for me i think i and MW have to live with a hanging MW from time to time.

I tried different BOINC versions too 6.4.5, 6.6.20 and the latest and only 6.6.20 gives usable behavior.

Is there a way to run two BOINC installations? - i think this would solve any BOINC problems.

I think best would be if we would assign projects to cores instead of hosts
ID: 25296 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 25302 - Posted: 13 Jun 2009, 20:15:12 UTC - in response to Message 25296.  

Is there a way to run two BOINC installations?

VM


ID: 25302 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MontagsMeeting

Send message
Joined: 12 Mar 09
Posts: 61
Credit: 9,214,340
RAC: 0
Message 25311 - Posted: 13 Jun 2009, 20:47:02 UTC

What is the performance impact of a vm?
i think i have to run the cpu-part on the vm or is the gpu available in a vm? that would be great
ID: 25311 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 25334 - Posted: 13 Jun 2009, 23:30:04 UTC - in response to Message 25311.  

What is the performance impact of a vm?
i think i have to run the cpu-part on the vm or is the gpu available in a vm? that would be great

I don't think you can run the GPU in VM, at least I haven't found a way of assigning the GPU hardware to it. But yes, you can then run CPU in VM to use your CPU to crunch MW as a well as the GPU crunching MW.

I'm not really sure about the performance impact. It's bound to impair overall performance since you have VM as an overhead. I can't really say there is any benefit since the number of WUs that can be crunched is pretty small compared to the turnaround that you can do with one GPU card, and you may also risk your system becoming unsuable for anything else with all that crunching going on.



ID: 25334 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 25359 - Posted: 14 Jun 2009, 1:09:07 UTC - in response to Message 25272.  

You should really set the avg_ncpus value higher. That way a lot less WUs are loaded and also the time-out issue is no longer present. If you loose one or two tasks is not much of an issue with a Core i7 because of the hyperthreading. And if you really want to avoid it use the cc_config.xml to raise the number of CPUs the BOINC client sees. If you set it to 9 and adjust your MW resource share to 11.1% (1/9) you can more or less reserve one of the 9 CPUs for MW and can afford a lot higher avg_ncpus values (you can even set it to 1.0). At least if you reset the debts it works quite well here.

But generally speaking the scheduler of the BOINC client is quite a mess in my opinion. The features of the newer clients (6.x.xx) like fractional CPU allocation are simply not working as they should.


Thank you for your helpful suggestions. I have used <avg_ncpus>1</avg_ncpus> and MilkyWay resource share of 12.5% with 8 BOINC cores and it has worked for the last 5 hours.

I tried using <ncpus>9</ncpus> and <zero_debts>1</zero_debts> in a cc_config file but MilkyWay processing times increased by 50% or more. 8 cores of Einstein do not share well with even a small amount of MilkyWay. I was previously running with only 7 BOINC cores to reduce excessive CPU heat, so didn't experience this before.

I don't mind using 1 core of 8 for speedy MilkyWay processing. This keeps my CPU load at the same 91% as using 7 BOINC cores. Also <avg_ncpus>1</avg_ncpus> means only 1 MilkyWay task in memory at the same time rather than 48 as I had with <avg_ncpus>0.01</avg_ncpus>.

The best thing is I don't have to restart BOINC evey hour to reset the timer on 48 MilkyWay tasks and there is no longer lots of tasks timing out when I am asleep.

Thank you and your testers very much for developing this application. It is great.


ID: 25359 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile KWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
Message 25554 - Posted: 15 Jun 2009, 19:00:27 UTC
Last modified: 15 Jun 2009, 19:04:07 UTC

OK, I give up. What the hell do I need to do to get MW from starting every single task. I'm using BOINC 6.4.7 and 0.19f. It's only processing 3 W/U's at a time but it shows all as running and at high priority to boot. All my debt is at 0. I just reset it hoping that would help but there was no change. It's been doing this for a long time now and I can't figure out what I need to do to stop it.
<edit>
My resource share is set equal for three projects. Prime Grid, MW, MW_GPU.
My cache is set to 1.5 days

4870 GPU
4870 GPU
ID: 25554 · 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 25557 - Posted: 15 Jun 2009, 19:23:25 UTC - in response to Message 25554.  

Mw will almost always run as high priority with the 3 day deadline. I can't recall about the solution for them all running, but I do remember seeing it here.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 25557 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MontagsMeeting

Send message
Joined: 12 Mar 09
Posts: 61
Credit: 9,214,340
RAC: 0
Message 25567 - Posted: 15 Jun 2009, 20:24:15 UTC

I believe there will be no real solution until BOINC allows us control our own hardware.
For me it looks like the devs are trying to get control over a million different systems running a million different combinations of projects in - guess what - a million different environments with another million goals of their users. I'm sure even to try this, is very stupid!
BOINC should do the basics and let the experienced and interested users enough room to configure BOINC to their needs.
I believe the complete ressource share concept doesn't fit todays hardware. It was OK for single core systems and worked so so on multi-core systems with equal cores/cpus but with the first P4 HT it was obsolete as HT could have a great performance boost if it would be able to assign projects with different cpu-usage to hardware-core and emulated-core. This problem growth the more different the hardware becomes. Actually we have Cores, HT-Cores, cuda, ati and we will become openCL, Larrabee (that perhaps can run cpu-apps and gpu-apps) someone mentioned somewhere soundcards and in worst case all in one PC - as i said, stupid idea to try to control such a system with a standard app - this will be hard enough with very detailed knowledge about the system.

In my eyes BOINCs goals are against their users, mainly the enthusiast users and if i look at the people i personally know and are running BOINC i only know one who wasn't influenced by me, there are five who run BOINC because i do the work configuring/babysitting their systems and another three i told to run it.

If BOINC is continuing to work against me, the point will be reached where i have to say good bye to BOINC taking at least five users with me but rather nine or ten. I would say people like me engage others to join BOINC and making us unhappy isn't a good idea.
ID: 25567 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 25586 - Posted: 15 Jun 2009, 21:45:43 UTC - in response to Message 25554.  

OK, I give up. What the hell do I need to do to get MW from starting every single task. I'm using BOINC 6.4.7 and 0.19f.

Use BOINC 5.10.45. That gives me 2 MW WUs crunching, the rest are either Waiting to start, or Waiting to report.

I have the same problem as you with BOINC 6.4.6 and 6.6.28


ID: 25586 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Rick Painter

Send message
Joined: 2 Mar 09
Posts: 3
Credit: 8,544,464
RAC: 0
Message 25598 - Posted: 15 Jun 2009, 23:40:45 UTC - in response to Message 25554.  

I ran BOINC 6.4.7 for ages and saw exactly the same MW high priority behaviour. I'm now running 6.6.31 and while it has many querks, I have not seen the high priority behaviour. I don't know if that has to do with the new BOINC version or I've configured it different or the stars are in the right position.

I started this thread trying to get SETI crunching on three cores and and one core supporting the gpu crunching MW. I got that by simply setting milky way to 25% resource share, SETI 75% resource share and <avg_ncpus>0.5</avg_ncpus>. Two MW crunching with no others running, three SETI crunching. Sweet!

The problem is that this is a very unstable situation. I've had SETI fill up the work cache which caused MW to stop calling for more work (solution was to disable SETI from calling work until it's really low and make sure it only fills up somewhat).

During a network outage last night I couldn't crunch MW for a while and BOINC carried on running with just SETI which was fine. However, when the network came back BOINC downloaded more MW but kept on crunching all SETI. Restarting BOINC didn't work, zeroing the debts didn't work, I finally suspended all SETI but three units to force BOINC to start on MW.

I agree with others on this thread about the bad behaviour of the scheduler and don't get me started about the debt system. I wish it was possible to switch that off. The amount of times I've seen BOINC bouncing around between applications, starting and stopping stuff when all I want it to do is to run application X on some cores and application Y on others with no change.

Anyway, enough ranting.
ID: 25598 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile KWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
Message 25790 - Posted: 17 Jun 2009, 13:29:40 UTC

Maybe if the BOINC Devs could put a toggle in the program allowing advanced users to manage things on their own and less advanced users to let the program manage things for them.
Just an idea.

4870 GPU
4870 GPU
ID: 25790 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 25794 - Posted: 17 Jun 2009, 14:23:37 UTC - in response to Message 25790.  

Maybe if the BOINC Devs could put a toggle in the program allowing advanced users to manage things on their own and less advanced users to let the program manage things for them.
Just an idea.

It has been suggested and rejected... you see, they know better how BOINC should operate than you ...
ID: 25794 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 25800 - Posted: 17 Jun 2009, 15:29:40 UTC - in response to Message 25794.  

Maybe if the BOINC Devs could put a toggle in the program allowing advanced users to manage things on their own and less advanced users to let the program manage things for them.
Just an idea.

It has been suggested and rejected... you see, they know better how BOINC should operate than you ...


They may know better but that gives no lee way for confirmed crunchaholics to manage their systems.

It might be useful to remove the lines referring to debt and watch the system crash.
Go away, I was asleep


ID: 25800 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MontagsMeeting

Send message
Joined: 12 Mar 09
Posts: 61
Credit: 9,214,340
RAC: 0
Message 25810 - Posted: 17 Jun 2009, 17:16:54 UTC

If i take control away from BOINC and babysit my PCs very intensely i can do about 180k credit per day. My actual RAC is about 130k, that means BOINC loses 30% and i still have to babysit. If i let BOINC manage things i hardly believe i could keep half of my RAC.
...they know better how BOINC should operate...
Yes, they know it should run half as fast and they do a great job >:D
ID: 25810 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Crunch3r
Volunteer developer
Avatar

Send message
Joined: 17 Feb 08
Posts: 363
Credit: 258,227,990
RAC: 0
Message 25819 - Posted: 17 Jun 2009, 18:54:26 UTC - in response to Message 25810.  

If i take control away from BOINC and babysit my PCs very intensely i can do about 180k credit per day. My actual RAC is about 130k, that means BOINC loses 30% and i still have to babysit. If i let BOINC manage things i hardly believe i could keep half of my RAC.
...they know better how BOINC should operate...
Yes, they know it should run half as fast and they do a great job >:D


You all seem to miss the priorities of DA regarding boinc development...

1. eye candy for the message board
2. social networking
3. add more bugs
.
.
.
.
9997. add more eye candy
9998. fix bugs and make boinc a usable tool.




Join Support science! Joinc Team BOINC United now!
ID: 25819 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · Next

Message boards : Number crunching : Configuring BOINC

©2024 Astroinformatics Group