| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: How can I limit WCG to 2 tasks on a 4 core processor? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 19
|
|
| Author |
|
|
Art_Brown
Cruncher Joined: Aug 24, 2007 Post Count: 7 Status: Offline Project Badges: 
|
I am running an Intel Core i7 920, 4 cores, single threaded, 6 GB RAM, and need to limit WCG tasks to 2.
How can I do that? In addition to WCG, I am running 2 GPUGRID tasks on 2 Nvidia GPUs, and one Test4Theory@Home task on one of the Intel cores. Running more than 2 WCG tasks causes the GPUGRID tasks to run at less than full GPU processor capacity. That is the problem I am trying to solve. Thanks for any suggestions. Art |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1412 Status: Offline Project Badges: 
|
Set "On multiprocessors, use: ... % of processors" to 75% and
keep the Test4Theory@Home resource share as high as or higher than WCG. |
||
|
|
Art_Brown
Cruncher Joined: Aug 24, 2007 Post Count: 7 Status: Offline Project Badges:
|
Thanks for the suggestion. I am running Boinc 6.12.34 for Windows 64 and have the Multiprocessor settings at "On multiprocessors, at most use no more than 1", as well as "On multiprocessors, use 10% of processors". I understand the latter setting is not used by this version of Boinc, however. The WCG Project Weight (share) is set to 10 and the Test4Theory Resource Share is set to 200. With these settings, 2 WCG tasks each take 25% of the total available CPU time, that is 100% of one core for each. Adding a 3rd WCG task takes about another 22% of the CPU capacity, and the GPU utilization drops, which is what I need to prevent.
Even though I issue the Update command to WCG, and even though I can verify the WCG settings on the WCG web site, it seems as though the settings are not being used. Is there any way to verify the values Boinc is actually using for WCG tasks? Any other suggestions? Thanks, Art |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
That wont work. The lowest of the 2 is 10% and that rounds to zero, so you'll need as per Crystal Palace 75% for 3 cores.
----------------------------------------It's in fact the 1 processor setting being ignored in 6.12. You need to clear local prefs [a button left top of local prefs screen] or use them exclusively. At any rate, if there is an ''override'' entry [in your message log], you're on local preferences. There's really no automated control over how many of each science run. The share game and low cache [0.0] and 'connect every xx days' to 0.00 that prevents there to be more WU's on the machine than cores in hopes that T4T get first work call. This complex selection of 2 WCG, 1 VM Test4Theory and 2 GPU jobs [that are impacted by --//-- edit: 2 clarifications. [Edit 2 times, last edit by Former Member at Sep 26, 2011 1:55:47 PM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
I am running an Intel Core i7 920, 4 cores, single threaded, 6 GB RAM, and need to limit WCG tasks to 2. How can I do that? In addition to WCG, I am running 2 GPUGRID tasks on 2 Nvidia GPUs, and one Test4Theory@Home task on one of the Intel cores. Running more than 2 WCG tasks causes the GPUGRID tasks to run at less than full GPU processor capacity. That is the problem I am trying to solve. Thanks for any suggestions. Art Hmm, if I'm understanding you correctly, you can share the 2 GPUGRID-tasks on a single core without any performance-hit, but if you're using more than 3 cores for cpu-tasks you'll get a performance-hit? If so, with the mix of projects you'll want, you can do: 1: Set "Use at most 75% of cpu's". 2: Set WCG's resource-share to zero. 3: Use a non-zero resource-share for Test4Theory and GPUGRID. With WCG having resource-share zero, you'll only ask for more WCG-work in case of idle cpu, meaning normally you'll only have two WCG-tasks and a running Test4Theory-task alongside the 2 GPU-tasks. But, since you can't get a new Test4Theory before the current one is uploaded, there is a possibility you'll get a 3rd. WCG-task while this upload is happening. In most instanses this 3rd. WCG-task will run for 60 minutes, your "Switch between applications every N minutes"-setting, so decreasing this to 1 minute will hopefully only wait 1 minute before the newly-downloaded Test4Theory-task starts. Now, one obvious disadvantage with setting WCG to zero resource-share is, you won't have any cache of cpu-tasks, so every time a task is finished chances are you'll sit idle for some seconds.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Really don't know as I've not seen the actual check-in notes from the developers (cant locate any). As posted by MarkJ: 6.12.36 change log (not released) - Mac installer: Bug fixes for OS 10.7 Lion. - client (and API and manager): change the data structure used for system process info Old: vector of PROCINFO. Descendants of a process were found by recursively iterating through the vector. Operations are O(n) New: map of (id, OCINFO), and each PROCINFO has a vector of its children. Operations are O(log(n)) Also combined Mac/Win/Linux variants of code that was essentially the same. - more code shuffling: proc_control: controlling processes procinfo: enumerating and querying processes run_app_windows: launching apps as other users on Win - client/API: add a mechanism so that apps can report sub-processes that are not descendants (e.g., virtual machines) These processes are then counted as part of the app, not as "non-BOINC CPU time". This fixes a bug where processing was incorrectly suspended because CPU usage by VM apps exceeded the "CPU usage limit" pref. Implementation: the PIDs of the processes in question are passed from app to client via shared-memory, in the app_status channel. A new variant of boinc_report_app_status() supports this. the VBox wrapper queries the PID of the VM, and reports it in this way. procinfo_app() includes a new argument: a list of PIDs that are part of the app, although not ancestrally related to the main process. in the client, ACTIVE_TASK now includes a vector "other_pids". If this is nonempty, it's passed to procinfo_app(). 6.12.37 change log - mac: add proc_control.cpp to XCode project. - lib: if OS_Darwin use procinfo_mac.cpp instead of procinfo_unix.cpp. - lib, api: update XCode project to build same source files in each library as corresponding make files. (Checked into 6.12.36 tag) - client: include the above in enforcing network quota preferences for virtual machine wrappers. 6.12.38 change log - client: Update Windows detection code to support Windows 8.(From: [P3D] Crashtest) - client: Disable GPU detection of Windows when running as a service to avoid a deadlock condition with the latest ATI drivers. "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
Art, go into the Bios and enable HT.
----------------------------------------In tools, Computer Preferences (locally), set "On multiprocessor systems, use at most 99% of the processor" The windows GPU app does not require much CPU time, or exclusive access (unlike Linux) but does need to have reasonable access. Freeing up one thread is sufficient, rather than a full core (even for both GPU's). This should reduce GPU task times and increase overall CPU throughput. The only way to specifically stipulate to run 2 WCG tasks would be to exclusively run CEP2 tasks, set to run 2 tasks, and keep a very small cache, with a very high WCG weighting. If you did this then the other project would fill the other slots. PS. 6.12.38 excludes GPU's and must be run as a daemon under Win8, and Art, mixing distinct GPU series architectures is not recommended. [Edit 1 times, last edit by skgiven at Sep 26, 2011 5:41:32 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Really don't know as I've not seen the actual check-in notes from the developers (cant locate any). As posted by MarkJ: 6.12.36 change log (not released) - Mac installer: Bug fixes for OS 10.7 Lion. - client (and API and manager): change the data structure used for system process info Old: vector of PROCINFO. Descendants of a process were found by recursively iterating through the vector. Operations are O(n) New: map of (id, OCINFO), and each PROCINFO has a vector of its children. Operations are O(log(n)) Also combined Mac/Win/Linux variants of code that was essentially the same. - more code shuffling: proc_control: controlling processes procinfo: enumerating and querying processes run_app_windows: launching apps as other users on Win - client/API: add a mechanism so that apps can report sub-processes that are not descendants (e.g., virtual machines) These processes are then counted as part of the app, not as "non-BOINC CPU time". This fixes a bug where processing was incorrectly suspended because CPU usage by VM apps exceeded the "CPU usage limit" pref. Implementation: the PIDs of the processes in question are passed from app to client via shared-memory, in the app_status channel. A new variant of boinc_report_app_status() supports this. the VBox wrapper queries the PID of the VM, and reports it in this way. procinfo_app() includes a new argument: a list of PIDs that are part of the app, although not ancestrally related to the main process. in the client, ACTIVE_TASK now includes a vector "other_pids". If this is nonempty, it's passed to procinfo_app(). 6.12.37 change log - mac: add proc_control.cpp to XCode project. - lib: if OS_Darwin use procinfo_mac.cpp instead of procinfo_unix.cpp. - lib, api: update XCode project to build same source files in each library as corresponding make files. (Checked into 6.12.36 tag) - client: include the above in enforcing network quota preferences for virtual machine wrappers. 6.12.38 change log - client: Update Windows detection code to support Windows 8.(From: [P3D] Crashtest) - client: Disable GPU detection of Windows when running as a service to avoid a deadlock condition with the latest ATI drivers. Yes, the semi proficient can find those, but was talking about 'the' check in notes at http://boinc.berkeley.edu/trac/browser/trunk/boinc/ P.S. Like the Zero Weight assignment to WCG. Few seconds idle... not so sure with 6.12, but if the line is up and that download delay is kicked stopped during upload (for the hardy hardy _4 file of CEP2, it would be moments. Sort of understood, though it's not the observation, that work requests were going to be postponed during upload. --//-- [Edit 1 times, last edit by Former Member at Sep 26, 2011 5:52:40 PM] |
||
|
|
Art_Brown
Cruncher Joined: Aug 24, 2007 Post Count: 7 Status: Offline Project Badges:
|
Thanks to everyone for their help. The current configuration I am running limits WCG tasks to 2, which is what I needed to achieve. The 2 GPUGRID tasks now use the 2 GPUs at their full capacity, and the system operates automatically without having to manually suspend any WCG tasks, as I had to do before. Also the Test4Theory task runs as it should. The Boinc Preferences settings are made in BoincTasks and are:
- Use at most 75% of processors (single thread per core on a 4-core CPU) - Switch between aplications every 1 minute - Use at most 100% CPU time On the WCG web site I set the rescouce share to 0.1. On the Test4Theory web site I set the rescource share to 900. On the GPUGRID web site I set the rescource share to 900, but I am not sure if this is necessary. With these settings, only 2 WCG work units are downloaded at any one time. The overall CPU load is about 85%. Total computer power consumption is about 600 Watts. The Intel Core i7 920 is overclocked to 3310 MHz and is stable. Triple channel DDR DRAM frequency is 919 MHz. I'll continue to try other configurations, including the Hyperthread setting suggestion. If you have any other suggestions, please post them and I will give them a try. Again thanks for your expertice! Art |
||
|
|
mfbabb2
Senior Cruncher USA Joined: Feb 18, 2011 Post Count: 361 Status: Offline Project Badges: 
|
- Switch between aplications every 1 minute
----------------------------------------Will be spending a lot of CPU overhead...
Murphy
 |
||
|
|
|