| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Varying success rates with different machines (RESOLVED) |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 19
|
|
| Author |
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges: 
|
I want to ask about differences in success between two computers.
----------------------------------------Last November, I installed BOINC on a new computer running Win 7 Home Premium 64 bit, with an Intel i7-920 four core processor and 6 gigs of DDR3 DRAM. The processor is hyper threaded, so eight threads running. I am set to using 100% of the processors at 60%. Starting with the then current build for this machine, and all default prefs, I had a heck of a time keeping work on this machine. It took about a month to get it sorted out. I tried a number of BOINC builds, I had a great deal of help from BOINC and WCG CA’s. Finally, what worked was to set Additional work at 2 days. I am crunching all of WCG and about eight other projects running BOINC software. I went from around 1000-1500 credits per day to about 5200 as of today. I saw very little in the way of complaints about the newer more powerful machines in either BOINC Messages Boards or WCG forums. It almost felt as if I was alone in having these problems. Since then, I have added another machine with an Intel i5-520M dual core processor, hyper threaded, so four threads, four gigs of DDR3 DRAM. This machine is running 100% of processors at 60%. In order to keep this machine in work, I had to add Additional work = 1.0. This machine is running all of WCG and 5 other projects using BOINC software. Now, my friend buys an even heavier weight machine than the one with the i7-920, running Win 7 Professional, a six-core Intel processor hyper threaded, 12 gigs of the most current DRAM. He only crunches on WCG projects so I do not believe that GPU is involved. He is using standard prefs, except I think he has the CPU usage set at 100% and 100%. He says his default Connect every = 2.5. He did not do any Additional work, which stands at 0. His credits per day have shot up to right now about 9000. He is having no problems keeping the machine in work. So, my question is, how can we determine what else might be at work in the success or failure of machines in keeping them furnished with enough work to not run out of work? (this question has also been posted in the BOINC Message Boards in Questions and Problems.) ---------------------------------------- [Edit 1 times, last edit by mitrichr at May 9, 2010 7:56:13 PM] |
||
|
|
pirogue
Veteran Cruncher USA Joined: Dec 8, 2008 Post Count: 685 Status: Offline Project Badges: 
|
Your chosen projects and any error messages from your log may be of help.
----------------------------------------Just out of curiosity, why the 100%/60% setting? I have mine all set to 100%/100% and I never really even notice BOINC running. I would guess that 100% usage of 50% (dual core) or 75% (quad core) of the CPUs would cut down on the task switching and might generate more points. |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
First, I miss-stated the Connect every on the new 6 core machine. I said it was 2.5. It is, of course, .25.
----------------------------------------Second, I have no current problem. My problems were all in the first month. I had great help, finally resolving the issues by using Additional work = 2.0. Last, I am at 100%/60% because that is the default. I want to change things as little as possible from the defaults. The question I raised is the really dramatic difference between the original success on my friend's six core machine with no real changes in the defaults, and the original miserable failure on my four core machine before changing prefs. That is the question. My friend runs only all of WCG, I run all of WCG, Einstein, Docking, Leiden, Rosetta, SETI, aqnd Spinhenge. Thanks. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello mitrichr,
This is a question that might differ according to some technical changes in the BOINC Scheduler, which I do not bother to keep up with. So I am just guessing here. First off, half your work is running as a second hyperthread on one core, so the scheduler might be a bit confused about the speed of your CPU. Particularly since it is trying to run only 3 seconds out of 5. I do not know if it understands about virtual cores, so I do not know if it is turning off everything for 2 seconds or individual threads. If everything ran at 100%, I expect that after a while it would predict the duration required by the work fairly accurately. Then it would be enough to ask for additional cache of 2.4 hours to make sure that a 2.4 hour communication window would keep the machine busy. This is just an ignorant guess based on my exasperation with the way the scheduler has misbehaved with predictions over the years. I try to give the scheduler the simplest possible situation, and even then expect some puzzling problems.  Lawrence |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
Hey Lawrence-
----------------------------------------How are you? The whole question is, my friend's machine, vastly more capable than mine, is yet way less demanding. His 6 core machine is also hyper threaded, and is doing much more work and in much less time than is my 4 core hyper threaded machine. Yet, with basically standard prefs, except for CPU usage of 100%, which makes the difference even more dramatic, he is not running out of work, while I was constantly running out of work until I went to Additional work = 2.0. >>RSM |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
My friend runs only all of WCG, I run all of WCG, Einstein, Docking, Leiden, Rosetta, SETI, aqnd Spinhenge.
time share between projects, sounds like you are not out of work, just out of wcg work, system busy with other projects |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
fredski-
----------------------------------------No, not out of work, I never said I was out of work. What I am asking is why the 6 core hyper threaded machine seems to be way less demanding, running on default prefs except CPU usage at 100%. On my 4 core hyper threaded machine, I needed to add Additional work = 2.0 to keep the machine working. I have plenty of work for WCG and all other projects. This comparison is making it sound like my 4 core machine is running through work faster than his 6 core machine. That is about which I am asking. |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
fredski- No, not out of work, I never said I was out of work. What I am asking is why the 6 core hyper threaded machine seems to be way less demanding, running on default prefs except CPU usage at 100%. The BOINC-defaults is 100%, but for some reason WCG has choosen to use 60%. For one thing this isn't a good solution, since this means cpu will run for 3 seconds and be idle for 2 seconds. Also, the way the throttling happens, it is a possibility application doesn't "catch" the off/on-signal, and get confused. Cutting-down #cores is a better solution than the constant on/off, if you really wants to cut-down at all... Since this isn't a laptop, there shouldn't really be any reason for not running at 100%. In case some programs interferes with this, you can either pause while active, or pause only when the particular programs is executing. But anyway, even with this constant throttling, this shouldn't have any effects on scheduling of work... On my 4 core hyper threaded machine, I needed to add Additional work = 2.0 to keep the machine working. I have plenty of work for WCG and all other projects. This comparison is making it sound like my 4 core machine is running through work faster than his 6 core machine. That is about which I am asking. It is a big difference between running a single project and running multiple projects. For single-project, the re-filling with new work is really not much difference between v4.2x and v6.10.xx. For multi-project on the other hand there's been multiple major re-writes through the years. Some of these re-writes has introduced bugs, that has been fixed in later versions. So, maybe you was really unlucky with your version, if not mis-remembers you've been running v6.10.1x? There's been made various changes later-on, so it's possible you wouldn't have ran-into similar problems with v6.10.5x that you had with v6.10.1x... For single-project user on the other hand, even if there is a bug in work-request, most of these bugs has no effect for single-project users.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
mitrichr,
----------------------------------------Don't have yourself confused about "this is no good at WCG or that is no good as the science app might get upset"... it's tested! Think the guys at WCG know very darn well what they're doing and have for the purpose of getting NEW crunchers to start set up 4 Pre-Set profiles. The default Standard Settings profile has 60%, mainly to prevent overheating of laptops for new arrivals... a thermal control. Regrettably the throttle is not very refined so it goes in a noticeable on/off pace and for that Fred developed the TThrottle utility to have BOINC run in smooth but lower CPU percent by setting the temperature ceiling... it's still on/off, just not visible to the naked eye. Now, if you're loath of working the preferences you can switch to the Maximum Output profile in My Grid > Device Manager. The default as said is Standard Settings. Click the Selected Profile link next to the device such as Default, School, Home, Work, and change the selection to the pre-set of choice, then save. On next client contact with servers that profile gets copied over. Then to have it all take effect, go into the In the FAQ's you'll find the differences between the different profiles if interested to know, but the titles of the 5 available are pretty much self explanatory what their objective is. As of version 6.10.33 or so the client introduces another System Wide load back-off control. That's default set to 25%. Many who are dyed crunchers have commented on that default level, including me, but as it is, it's made for new start crunchers where it's wanted that they truly do not notice crunching. If they want to contribute more than it's RTFM time ... a long read ;D Happy crunching.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at May 8, 2010 1:18:02 PM] |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
Hey Sek - Nice to hear from you.
----------------------------------------You remember me back in October or November with the i7-920? Those were miserable days. Could not get anything to keep the machine in work. Then, over the US Thanksgiving holiday, off to California to see the kids, but still trying to figure things out. When I came back, that was when someone suggested the Additional work X=>0. I tried 2 days and that solved everything, leaving all other prefs alone. So, since then I am a happy camper. I check the projects just to be sure that I am not having so much work that it is getting done late. I am leaving everything as it stands. My surprise is really on two levels: I saw very few if any other people with a machine of my caliber, i7-920 a fair amount of DRAM, etc., having any difficulty. And, then, my friend gets a machine which should blow mine out of the water and he on default prefs has a nice list of WU's ahead of him, maybe 40 or so. If anything, his CPU usage of 100% should make his machine even faster. This last is what prompted my question. All five of my machines are doing nicely, not running out of work and not getting work done late. Hope all is well. >>RSM |
||
|
|
|