| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Windows 7_64 Uitimate and BOINC |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 58
|
|
| Author |
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
I just started a week ago to run Windows 7 64bit Home Premium on an I7-920 CPU. So, with hyperthreading, I would run 8 WU's simultaneously. I first installed 6.10.18, got a bumper crop of WU's from about a dozen projects to which the machine is attached. But, as they finished, they were not replaced. So, I installed 6.6.38, same result. New install, or upgrade of older client? If upgrade, likely reason is the LTD getting highly negative by the upgrade, it's possible to reset it immediately. Or, you can just let it continue to run. If new install, your cache-settings is most likely the reason. Also please note, while old BOINC-clients will normally try to keep a full cache in all projects with positive LTD, as long as isn't in deadline-trouble, in v6.6.xx and later clients the cache is total across all projects. So, if you're example using a cache-setting of 0.1 days, if you've got 1 running task in 8 projects and maybe 1 extra in project-#9, it's a good chance your cache is full, and it will be re-filled when needed. Is there any data good or bad on Win 7 64bit on an I7 CPU? No problems here, but I'm still running the win7-RC...  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges: 
|
Ingleside-
----------------------------------------Thanks for your questions. I am not technical, what is LTD? The computer is brand new as of last week, and I bought it with Win 7 64bit Home Premium. I went to BOINC to look for what to install. I started with the latest, Nov 9, 6.10.18. Then I went back uninstalled 6.10.18, rebooted and installed 6.6.38 from late October.. Finally, I repeated the steps and installed 5.8.16 which is from 2007 and still kicking around, and so I would think quite solid. I have been taking screen shots and saving them so that I can see if new WU's are coming. Of course, I am also checking messages about this issue. I am seeing a trickle of new WU's, and I am no where near out of work - yet. >>RSM |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
I found what is LTD.
---------------------------------------->>RSM |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
mitrichr
----------------------------------------If you go back to your discussions at BOINC and find posts v.v. the number exceeding -86400, work is no longer fetched. The quick fix is to reduce cache/buffer to zero, set the connect interval to 0.01 days in the local prefs, detach from all the projects and reattach in the order of your most important projects to the least. Stops the initial result cache flooding and allows for the projects to find their ground. Only then slowly increase cache, each day a bit until the client learns who does what how and when. Never exceed your cache size above the shortest project deadline or you will near guaranteed be in for future trouble and project run time/weight disparity.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
O.K., thanks, where do I find the cache/buffer in 5.8.16, where I find very little in the way of configuration settings? In fact, I do not even see a setting for connect interval.
----------------------------------------I looked on a machine running 6.2.28, where I do of course see a setting for connect interval; but again, nothing that says cache/buffer or either word alone. I assume since you suggest that I use these controls that I should install a version that does in fact have these controls available, which the version I am running does not. It is, now, also running low on work and will peter out. I regret that my lack of technical skills is such a burden to you all; but this work is very important to me, the whole reason that I bought this machine while having two perfectly good and not too old Vista dual core machines. >>RSM |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
O.K., thanks, where do I find the cache/buffer in 5.8.16, where I find very little in the way of configuration settings? In fact, I do not even see a setting for connect interval. The only reason v5.8.16 is still available on the download-page is that this is the last version that works with win95 and NT4. For other OS like XP, Vista and Win7, it's better to use a newer version. v5.8.16 missing many features, is more likely to give errors (especially on Vista and win7), and also bigger chance of only asking for work if idle cpu... I looked on a machine running 6.2.28, where I do of course see a setting for connect interval; but again, nothing that says cache/buffer or either word alone. When looking at the BOINC Manager local preferences, the cache-size is the sum of "Connect about every N days" and "Additional work buffer N days". If you've got a permanent connection, it's recommended to set "Connect..." to zero, and use "Additional..." instead, example start with 0.1 days for "Additional..." and increase if needed/desirable. I assume since you suggest that I use these controls that I should install a version that does in fact have these controls available, which the version I am running does not. It is, now, also running low on work and will peter out. Using v6.10.xx has some advantages when it comes to tracking-down problems than earlier versions... For one thing, in advanced view, Projects-tab, when you selects an individual project, you can by hitting "Properties" get info about why a particular project isn't asking for more work. Example, "CPU work fetch priority" = LTD. It also tells you if a project currently is being deferred. In v6.10.xx, there's two rules for when projects is asked for work (if more work is neccessary), depending on the two cache-settings "Connect..." and "Additional...": 1: Always try keeping "Connect..." work on client, this means asking all contactable projects for work, starting with the projects with zero LTD, and working to more and more negative. 2: For the "Additional..." on the other hand, only ask projects for work if -LTD < "Connect..." + "Additional...", except if idle cpu. So, example, if "Connect..." is zero and "Additional..." is 0.5 days, on an 8-way i7 this means any LTD < -345600 gives no work-request to this project, except idle cpu. I regret that my lack of technical skills is such a burden to you all; but this work is very important to me, the whole reason that I bought this machine while having two perfectly good and not too old Vista dual core machines. >>RSM No problem.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
So, O.K., I installed 6.10.18.
----------------------------------------I set Connect to 0 and Additional to 0.10 Over some time, I attached to four projects. One project got immediately six WU's, the other three got immediately only one each. Now here are the figures from Properties: CPU sched priority -22266.51 CPU work fetch priority -50834.66 Deferred ------- CPU sched priority 7285.90 CPU work fetch priority -23.40 Deferred not shown as a category at all CPU sched priority 307.44 CPU work fetch priority 00.00 Deferred ------- CPU sched priority 14981.00 CPU work fetch priority -10445.45 Deferred ------- So, those are the figures at a point in time over about the three-four minutes it took me to write them down. I needed to go back and check Deferral, and I saw that the figures change. I do not know the math that got you to -345600, so I do not know how to figure for LTD for any setting. So, I don't really have any idea at all where I am. Whoops, ignorant yes, but it looks like the setting of 0.10 total worked somewhat, I just got some new WU's. I have eight running and four Ready to Start. That is a dynamic - change - I welcome. If you can explain to me the math and what response to make to the math, it might be very helpful. >>RSM |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
So, O.K., I installed 6.10.18. I set Connect to 0 and Additional to 0.10 Over some time, I attached to four projects. One project got immediately six WU's, the other three got immediately only one each. Now here are the figures from Properties: #1: CPU sched priority -22266.51 CPU work fetch priority -50834.66 Deferred ------- #2: CPU sched priority 7285.90 CPU work fetch priority -23.40 Deferred not shown as a category at all #3: CPU sched priority 307.44 CPU work fetch priority 00.00 Deferred ------- #4: CPU sched priority 14981.00 CPU work fetch priority -10445.45 Deferred ------- Based on these numbers, the projects will currently ask for work in the order: #3, #2, #4, #1, and all can ask for work with your current cache-settings. For order of running CPU, it's currently #4, #2, #3, #1. As for "Deferred", on the Properties-page, there'll only be any numbers if you've asked a project for work, but the server has responded with "no work". There can also be showing "Deferred" on the Projects-tab, for one or more projects at once. This is either the mandatory delay since last connection to project (normally from 7 seconds to maybe 5 minutes, depending on project). Or, it's deferral in case you've got connection-problems. So, those are the figures at a point in time over about the three-four minutes it took me to write them down. I needed to go back and check Deferral, and I saw that the figures change. I do not know the math that got you to -345600, so I do not know how to figure for LTD for any setting. So, I don't really have any idea at all where I am. There are 86400 seconds in a day. (60 seconds * 60 minutes * 24 hours) You've got an 8-way system (i7 with ht). In my example I used a cache-setting of 0.5 days. This give the math: "seconds per day" * "#cores" * "cache-size" = 86400 * 8 * 0.5 = 345600. Since all LTD's is adjusted to max be zero, and otherwise negative, the cutoff work-request is for -345600, on an 8-way system with 0.5 days cache-setting. With your current 0.1 days cache-setting, the math is: 86400 * 8 * 0.1 = 69120. Since the project currently with lowest LTD, #1, is at -50834.66, all projects can ask for more work if needed, since -50834.66 > -69120. Whoops, ignorant yes, but it looks like the setting of 0.10 total worked somewhat, I just got some new WU's. I have eight running and four Ready to Start. That is a dynamic - change - I welcome. If you can explain to me the math and what response to make to the math, it might be very helpful. >>RSM As long as you're getting a steady stream with new work, you don't really need to do anything.  The numbers is mainly as an help in cases like example: 1: Why aren't project-X asking for work, it's currently out of work? 2: Why is the project-Y with the 0.01% resource-share constantly asking for more work? And so on... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 2 times, last edit by Ingleside at Nov 21, 2009 12:32:58 PM] |
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges:
|
Thanks so much for all of your help.
----------------------------------------After I reported new work to you, I went to bed. So, that meant a few hours to either finish work and get new work or finish work and be out of work. When I arose, I had a modest amount, 8 WU's working, a modest amount of work waiting to run and waiting to start. But there was nothing as long as even 6 hours. I left at 8:30AM to go cycling. I was gone from8:30AM-3:00PM , six and one half hours. I figured I would come home to either no work, a little work, or everything in good shape. I cannot be happier. Plenty of work, plenty waiting. Thanks so much for all of your help, and, especailly the math. I was surprised at the ranking you gave the four examples I pulled from Properties. I thought the higher the negatives, the greater the debt, the greater the need for work. But, hey, that's why you are you and I am me, and, I am better for having had your good counsel. >>RSM |
||
|
|
Somervillejudson@netscape.net
Veteran Cruncher USA Joined: May 16, 2008 Post Count: 1065 Status: Offline Project Badges: 
|
Sekerob thanks!
|
||
|
|
|