problem with core 2 duo not running both cores

Message boards : Number crunching : problem with core 2 duo not running both cores

To post messages, you must log in.

AuthorMessage
Profile Changeling

Send message
Joined: 6 Dec 05
Posts: 2
Credit: 866,179
RAC: 0
Message 31204 - Posted: 15 Nov 2006, 21:50:40 UTC

Hello all,

just got a new computer, going through the normal processes, delete junk, add needed programs, set up boinc...etc.

I'm running seti@home and rosetta@home split 50/50. I went to both preferences and switched to "work" after setting the use processor number to 4 (I didn't know if it had hyper threading) in both projects.

Everything was fine, 2 work units running at the same time, until sometime around noon today, when for no apparent reason it stopped running 2 units and started running just one. I got an over committed message, but if the computer was running both cores this wouldn't be a problem.

Anybody have any ideas? The comp is a core 2 duo 2.0ghz with 4 mb cache, 2 gig memory

Any help appreciated!!

Changeling
ID: 31204 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
MikeMarsUK

Send message
Joined: 15 Jan 06
Posts: 121
Credit: 2,637,872
RAC: 0
Message 31205 - Posted: 15 Nov 2006, 22:08:50 UTC

Once you get an overcommitted message, Boinc will stop downloading new workunits (regardless if that leaves the second core idle). If you increase the overcommitted project's resource share so that Boinc thinks the result will complete in time, then 'overcommitted' should go away and it'll then download new work.

ID: 31205 · 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 31214 - Posted: 16 Nov 2006, 2:18:06 UTC

If BOINC thinks it is behind for one of the projects, it should try to run both cores on two WUs of that project. So, what you are observing doesn't make much sense.

Perhaps answers to the following might turn up something:

What BOINC version?
How long was it running both cores after you set it up?
Do both projects have WUs ready to crunch on your machine?
Have you changed your Rosetta runtime preference since setting up this machine? ...or perhaps it is still adjusting to what appears to be your 8hr preference.

...I'll bet that's it. I think BOINC, just outta the box, assumes R@H work will take 3 hours, but you've got an 8hr preference... so it initially tried to grab enough work for your cache size based upon 3hrs per WU, and now BOINC is learning otherwise and feels it's got more work then it would like to have.

If Rosetta is the project that is still crunching, that might be what happened... but I'd STILL think it should be running TWO R@H WUs then.

Either way, I'll bet it straightens itself out within 24hrs crunch time.
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: 31214 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
zombie67 [MM]
Avatar

Send message
Joined: 11 Feb 06
Posts: 316
Credit: 6,410,102
RAC: 0
Message 31215 - Posted: 16 Nov 2006, 2:51:58 UTC - in response to Message 31214.  

...I'll bet that's it. I think BOINC, just outta the box, assumes R@H work will take 3 hours, but you've got an 8hr preference... so it initially tried to grab enough work for your cache size based upon 3hrs per WU, and now BOINC is learning otherwise and feels it's got more work then it would like to have.


Good point. R@H run time is very sensitive. I recommend making changes in increments of 1 hour (up or down) per day. Give BOINC time to work *with* your changes.
Reno, NV
Team: SETI.USA
ID: 31215 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile sslickerson

Send message
Joined: 14 Oct 05
Posts: 101
Credit: 578,497
RAC: 0
Message 31216 - Posted: 16 Nov 2006, 2:52:40 UTC

I've had to manually update Rosetta a few times today to get work for both cores. I haven't changed my preferences lately and I am running Rosetta on both.

Like everyone else I had many errors over the last few days, anyone think that could have anything to do with it? Although I just noticed I am back to 100 results/day normal.

Tim



ID: 31216 · 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 31217 - Posted: 16 Nov 2006, 5:26:46 UTC - in response to Message 31216.  

Tim (slick), with a RAC of 410 on that machine I presume it is crunching 100% Rosetta?

I'd think that the failed WUs would mislead BOINC into thinking work gets done faster then your preference. There is a value in your General Preferences for how often to connect to the network. The default is .1 days (2.4 hrs). So BOINC is basically assuming that it will have network access every 2.4 hrs. So it's only trying to get about 3hrs of work. Which would basically be two WUs for your dual core. So, if one of those WUs fails, or otherwise finishes early... well, BOINC's not smart enough (someone will correct me if I'm wrong or outdated here) to realize that attempting an earlier then planned connection to the network would be a great way to keep crunching.

Suggest you do two things.
1) Increase your preferred runtime. (it looks like perhaps you've got it set to 2 hrs?) I like Zombie's suggestion to bump it an hour a day. This is done in your Rosetta preferences.
2) Increase your General Preference for "connect to network about every ... days". Again, make gradual changes here as well. ...and changing both of these settings at the same time is a bit more confusing. So, I'd suggest doing #2 first to assure you've got more WUs on-deck. This helps assure you've always got work to crunch.

I've got mine set to 1.2 days, even though I basically have full time internet access. This carries me over adversity... such as the new 5.40 version coming out. This changes the application name that's trying to access the net, and caused my firewall to deny it access all night. But I was able to resolve the problem the following day and still have plenty of work to crunch. Also carries you if your internet access is... well, less then perfect. And gives you coverage even if the R@H servers are down for an upgrade or have an outage of any kind.

It also allows me to establish time of day for network usage limits. This allows me to do my network access in the dark of night, when there's no strain on the rest of the network.
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: 31217 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile sslickerson

Send message
Joined: 14 Oct 05
Posts: 101
Credit: 578,497
RAC: 0
Message 31254 - Posted: 16 Nov 2006, 15:17:54 UTC - in response to Message 31217.  

Tim (slick), with a RAC of 410 on that machine I presume it is crunching 100% Rosetta?

I'd think that the failed WUs would mislead BOINC into thinking work gets done faster then your preference. There is a value in your General Preferences for how often to connect to the network. The default is .1 days (2.4 hrs). So BOINC is basically assuming that it will have network access every 2.4 hrs. So it's only trying to get about 3hrs of work. Which would basically be two WUs for your dual core. So, if one of those WUs fails, or otherwise finishes early... well, BOINC's not smart enough (someone will correct me if I'm wrong or outdated here) to realize that attempting an earlier then planned connection to the network would be a great way to keep crunching.

Suggest you do two things.
1) Increase your preferred runtime. (it looks like perhaps you've got it set to 2 hrs?) I like Zombie's suggestion to bump it an hour a day. This is done in your Rosetta preferences.
2) Increase your General Preference for "connect to network about every ... days". Again, make gradual changes here as well. ...and changing both of these settings at the same time is a bit more confusing. So, I'd suggest doing #2 first to assure you've got more WUs on-deck. This helps assure you've always got work to crunch.

I've got mine set to 1.2 days, even though I basically have full time internet access. This carries me over adversity... such as the new 5.40 version coming out. This changes the application name that's trying to access the net, and caused my firewall to deny it access all night. But I was able to resolve the problem the following day and still have plenty of work to crunch. Also carries you if your internet access is... well, less then perfect. And gives you coverage even if the R@H servers are down for an upgrade or have an outage of any kind.

It also allows me to establish time of day for network usage limits. This allows me to do my network access in the dark of night, when there's no strain on the rest of the network.


Thanks!

I was actually thinking about doing those very two things. I have an always on connection (although not always working) but I prefer shorter run times just to satisfy my craving for new work. With so many different WU types its nice to do a little of each. I'll increase both slowly and see what I get...



ID: 31254 · 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 31257 - Posted: 16 Nov 2006, 16:21:07 UTC - in response to Message 31254.  

I prefer shorter run times just to satisfy my craving for new work. With so many different WU types its nice to do a little of each.


If you think about it, you'll STILL get all the variety of WUs... you'll just have them on deck for a half day or so before you crunch them. It's sorta like watching all the movies that come out, but waiting for them to come out on video :) ...but you have the advantage, if a problem is found with a given WU, you can read about it BEFORE you're machine is ready to crunch on it :) ...in other words... by waiting for movies on video, you have the advantage of reading more reviews to avoid bad movies with great success.
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: 31257 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile River~~
Avatar

Send message
Joined: 15 Dec 05
Posts: 761
Credit: 285,578
RAC: 0
Message 31262 - Posted: 16 Nov 2006, 20:09:22 UTC

I don't think it is anything to do with the box being overcommitted - I have had my 2-cpu boxes overcommitted and this has never stopped both cpus running. BOINC can't tell the difference between having two cpus in separate sockets, or two cpus in separate cores in the same socket.

What I do see sometimes is an annoying effect where the box suddenly thinks the number of available cpus has changed. Looking in the message tab, is there a benchmarks message? These are quite obvious as they go on for several lines.

One cause of this I have found is when you have different numbers of cpus set in different venues, and when the host is at a different venue in different projects. There is a funny interaction between the fact that each project can have a separate venue, and the fact that gen prefs come from only one project.

I avoid this problem (which I do not fully understand) by keeping the following advice

1. Always make changes to general prefs from the same project. In you case, choose either SETI or Rosetta as a "control" project and stick to it.

2. Make sure the host is set to the same venue in each project

3. Make sure the general prefs (as set in your control project) show 2 or more cpus for that venue

4. Make that venue the default venue for new machines (sometimes BOINC unilaterally decides your box is a new machine, and this can force a change in venue, also forcing a change in the number of cpu).

Please let us know if this advice helps.
River~~
ID: 31262 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile skgiven
Avatar

Send message
Joined: 7 Jun 06
Posts: 9
Credit: 642,742
RAC: 0
Message 31275 - Posted: 16 Nov 2006, 23:57:12 UTC - in response to Message 31204.  
Last modified: 17 Nov 2006, 0:00:13 UTC

Hello,

I have a similar problem with one of my servers. I have BOINC running and although it uses both cores it does not use them entirely. The processor never max’s out. The 2 cores hover around 50% each in Task Manager, Performance. Odd! It also seems to think there is one processor on the benchmark at times. Just gone past odd heading for weird!

2003 Server R2 Pentium D 945 (Two 3.4GHz cores OC to 3.78), 2GB 667, 600GB…
Any suggestions on how to get the most out of the Processors?


skgiven


ID: 31275 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile sslickerson

Send message
Joined: 14 Oct 05
Posts: 101
Credit: 578,497
RAC: 0
Message 31276 - Posted: 17 Nov 2006, 0:27:01 UTC - in response to Message 31275.  

Hello,

I have a similar problem with one of my servers. I have BOINC running and although it uses both cores it does not use them entirely. The processor never max’s out. The 2 cores hover around 50% each in Task Manager, Performance. Odd! It also seems to think there is one processor on the benchmark at times. Just gone past odd heading for weird!

2003 Server R2 Pentium D 945 (Two 3.4GHz cores OC to 3.78), 2GB 667, 600GB…
Any suggestions on how to get the most out of the Processors?


skgiven




All of your hosts appear to show the correct number of CPU's so I do not know why the benchmark is picking up only one CPU. Could you please let us know which host you are speaking of and possibly paste the Benchmark message from the BOINC client here.

As far as CPU usage, on a dual core machine you are seeing perfectly normal behavior. You'll notice that on Task Manager "Performance" CPU usage should always be at 100%. Each Rosetta process will display somewhere around 50 and will vary as other processes take CPU cycles. This is Normal.

Tim



ID: 31276 · 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 31279 - Posted: 17 Nov 2006, 3:40:18 UTC

sk which BOINC version are you running on the box in question? BOINC is adding a new feature, I believe it is in beta presently, which allows you to use a fraction of the CPU. Check General Preferences top section on processor usage, last entry for "use at most ... %". You want 100%.

Some users have machines that run too hot, or fan makes too much noise, so they want to use less then 100% to control things. BOINC goes idle for a second once and a while to achieve the desired CPU usage... but only once you get to... wish I could remember which BOINC version. I think it is BOINC 5.5.
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: 31279 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile KSMarksPsych
Avatar

Send message
Joined: 15 Oct 05
Posts: 199
Credit: 22,337
RAC: 0
Message 31332 - Posted: 18 Nov 2006, 0:16:45 UTC - in response to Message 31279.  

sk which BOINC version are you running on the box in question? BOINC is adding a new feature, I believe it is in beta presently, which allows you to use a fraction of the CPU. Check General Preferences top section on processor usage, last entry for "use at most ... %". You want 100%.

Some users have machines that run too hot, or fan makes too much noise, so they want to use less then 100% to control things. BOINC goes idle for a second once and a while to achieve the desired CPU usage... but only once you get to... wish I could remember which BOINC version. I think it is BOINC 5.5.



It was introduced in the 5.5 alpha line. They are alpha testing the 5.7 line.

Use at your own risk. I can tell you I'm having a heck of a time getting the throttling to work. It runs for about 5.5 minutes wall time and then all processing stops.
ID: 31332 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : problem with core 2 duo not running both cores



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