Message boards :
Number crunching :
Multi core greater than 80 core
Message board moderation
Previous · 1 . . . 4 · 5 · 6 · 7 · 8 · Next
Author | Message |
---|---|
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Maybe the problem BOINC faced is that when the first BOINC was made (~2005 ?) the description of the GetSystemInfo Function given by Microsoft maybe was: GetSystemInfo returns "The number of logical processors in the system" and now (when bigger systems exist and they invented "processor groups") Microsoft changed the description to: GetSystemInfo returns "The number of logical processors in the current group" If you have Microsoft compiler from 2005 (or older) can you check in its help files what is the description of the GetSystemInfo Function? Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
It says it in the old way - "Number of processors in the system.". MSDN from 2005, I'm still using VS 2005. Although misleading with its new behavior, it doesn't change anything. Old applications, which are not able to get right number of processor when using this function, are probably unaware of existence of nodes and how to optimize their thread grouping on different nodes anyway and they don't expect to run on that large systems at all. One of the rare cases when u need to know how many processors are on your disposal, is when creating I/O completion port and defining number of active threads. And even then old application would keep 128(its not fixed number, its just best having 2 threads per processor) active threads. But, note, they would be automatically scaled on other available processors, no matter how many processors this application is aware of. |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
It says it in the old way - "Number of processors in the system.". MSDN from 2005, I'm still using VS 2005. It is still misleading, and the default for the changed behaviour in Windows for backwards compatibility caused a problem for Boinc that has taken time to fix. Unfortunately, such problems are 'not unusual' in the Windows world. For (all?) other operating systems, perhaps 'greater care' is taken to maintain backwards compatibility or to give a drop-dead meaningful error message. Hence, things should work as intended, or, the program is stopped dead to force whatever problem due to 'whatever has changed' to be fixed. I'm sorry, but there is great suspicion when there is a very deliberate marketing policy to force people to continuously upgrade and pay. Sorry, your earlier response is meaningless. It looks like you don't want to usefully compare OS features. And you are blind to the computer science of schedulers. Please enjoy your ultimate can-never-be-improved OS. Meanwhile, at least Linux is fully open to all scrutiny and from that, to meaningful improvement. Good kudos to the Boinc development that the fix for Windows is being put in place as quickly as possible. Happy fast computin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
Thank you for ruining consequent thread with unnecessary Linux bloat. It is certainly not the way to convince Linux is better OS. I promise to report your posts every time they go even slightly out of topic from now on. |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
Thank you for ruining consequent thread with unnecessary Linux bloat. It is certainly not the way to convince Linux is better OS. Hardly. I was hoping you would offer your expertise to shed some light on the comparison of schedulers and explain how the Windows kernel scheduler is better or not and how the Windows Groups is the only way to go... And worth the extra programming effort. I'm genuinely interested in how Windows does it's scheduling. However, detailed information on the web is scant. Is it all secret and 'proprietary'? This issue gets all the more important as we move to ever more processor cores in systems. Happy fast crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
Read "Windows Internals" books, watch "Channel 9" videos om MSDN, read developer blogs and comments. Mark Russinovich and other guy from kernel team can enlighten you. Here is a good starting point and another one too If u dig few clicks deeper u may find this. |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
Read "Windows Internals", watch "Channel 9" videos om MSDN, read developer blogs and comments. Mark Russinovich and other guy from kernel team can enlighten you. 44 minutes 30 seconds video... Is there a "Wikipedia" style succinct description that any programmer can read please? I'll take a look the vid at some time. Multitasking on other things at the moment... Have you checked out the O(1) scheduler?... Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
Yes, its two paragraphs, also its too general. I read information which only concerns my current task. Life is too short, information is too much. And, tbh, Linux scheduler is very deep down in the queue. Sorry for not being good companion. My intention when participating this thread was to point out that things are neither so simple, nor so "dark", concerning Windows. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I've been wondering how people have been racking up the millions on dual core and four core machines when I'm just chugging along with my multiple 16+ core systems... I for one have a large number of idle machines on hand & 99.99999% of their time is spent processing S@H work. Depending on the workload of your 16 core machines & when they are allowed to run I would think they would do about 15,000 a day in RAC. The use of optimized apps increases the efficiency by making use of the advanced functions in the processors. If you want to go crazy then you put 4 dual processor video cards in a system & it cranks out several 10,000's worth of RAC a day. Some ever crack over 100,000. GPU computing while powerful seems to be "temperamental". With 8 & 10 core CPU's having 80 logical processors is a fun idea & hopefully getting them all going full tilt can be done. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
-BeNt- Send message Joined: 17 Oct 99 Posts: 1234 Credit: 10,116,112 RAC: 0 |
Thank you for ruining consequent thread with unnecessary Linux bloat. It is certainly not the way to convince Linux is better OS. Talking about schedulers, in a thread.....talking about using more of a machines cores in a Windows environment......I mean off topic is fine with me ever so often, but you bring your linux spew into EVERY thread where someone has an issue with something running on a Windows machine, where I might add often times the issue they are having is not the fault of Windows, but either user, or user contributed software problems. Speaking of which how did that turn out? I can't tell as it seems this has turned into yet another 'Linux is best / Windows sux' campaign. Traveling through space at ~67,000mph! |
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
At the moment, I have adjusted the config file to now read: <ncpus>40</ncpus> It was set to 80. I saw 80 running SETI tasks in Task Mangler. I also saw my cpu utilization drop down to ~37%. I thought this might have to do with using logical processors versus cores so I dropped it down to 40 (actual num of cores in this system...which btw has SMT enabled in BIOS...ergo the 80 logicals seen by the OS). Anyhoo, the system is now still showing 80 tasks running although BOINC shows from Read config file that "...use at most 40 cpus." That all said, when can I expect BOINC to stop running 80 tasks and scale back down to 40? Also, when it does, will my cpu utilization begin to scale back upward? I would hope so... BOINC seems slow to accept and implement changes to the config... |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Did you just do a 'Read config file'?, did you try restarting Boinc? Claggy |
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
I wasn't aware BOINC needed to be restarted to accept config file changes. I guess I hoped it was dynamic after a config file read...which one would be led to believe...otherwise why read the config file :) |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
I wasn't aware BOINC needed to be restarted to accept config file changes. I guess I hoped it was dynamic after a config file read...which one would be led to believe...otherwise why read the config file :) Have you got the following setting enabled?: Leave tasks in memory while suspended? yes If you have then 80 tasks will remain showing in Task Manager, 40 of them will be computing, and 40 of them will sit idle awaiting to continue when a running task completes, (they will be taking up memory) Now if you restart Boinc, only 40 tasks will show in Task Manager, the other 40 will show in Boinc Manager as waiting to run, and will restart from their last checkpoint, (as opposed to being sat in memory waiting to run.) Claggy |
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
Mine is set to "no". Changed my config file back to <ncpus>80</ncpus> and restarted BOINC. Now seeing all 80 tasks running in both BOINC Mangler and Task Mangler. Overall CPU usage has jumped up to ~ 75%. Now how to use up that final 25%...? |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Mine is set to "no" Is that with Web based Computing preferences? and in the same venue as your host is set too? are there any local preferences set on that host? (local preferences over-ride Web preferences) (I haven't seen any startup messages for that host, so sometimes difficult to visualise how things may be set up) Claggy |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
I wasn't aware BOINC needed to be restarted to accept config file changes. I guess I hoped it was dynamic after a config file read...which one would be led to believe...otherwise why read the config file :) http://boinc.berkeley.edu/wiki/Client_configuration http://www.boinc-wiki.info/Cc_config.xml http://boincfaq.mundayweb.com/index.php?language=1&view=91 I copied the "Warning" from the last link to the first two links (so hopefully no more "I wasn't aware" will happen): Warning: Any changes in the <log_flags> section will be used immediately after re-read of the config file. Yet a lot of changes in the <options> section will need a complete exit and restart of the BOINC client. They will not take if you just re-read the config file. (I think the info in "BOINC FAQ Service" is free to use) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Changed my config file back to <ncpus>80</ncpus> and restarted BOINC. Now seeing all 80 tasks running in both BOINC Mangler and Task Mangler. Overall CPU usage has jumped up to ~ 75%. Now how to use up that final 25%...? It is supposed to use all the available CPU time. 1) In Windows Task Manager - do you see every of the 80 SETI tasks using 1.25% CPU (1.25 * 80 = 100%)? Are there other processes using much CPU? 2) In BOINC Manager -> Tasks tab - are all the 80 SETI tasks show "Running" (or e.g. "Waiting for memory", "Waiting to run", etc...)? Is the percentage counting up in the "Progress" column for all the 80 SETI tasks? Restart BOINC and post the first 40-50 lines from the Messages tab (Click, Shift-Click, use [Copy selected messages] button at the left) These messages will help us see many settings which are in fact in use (some set by you, some defaults) - will save our and your time asking for every setting. Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Dave Send message Joined: 29 Mar 02 Posts: 778 Credit: 25,001,396 RAC: 0 |
So may processes sounds like more overhead than it's worth. |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
160 processes with 3 threads each, probably two of them in wait state, are not generating much overhead. More noticeable problem is when they initialize(OS initialization phase and later process initializing its internal data and structures), which happens at same time. Another bottleneck can be slow memory, bcs of the concurrent access, already in their main working state(so count it as hardware limitation). |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.