6.3.21 for Windows x86 seems OK x 2 installs.

First start had correct max 2 cpus + 1 CUDA/CPU<1

I ran several quick tests suspending a CPU task, another always started in it's place.

I did several quick tests changing cpus from 100% for 2, 50% for 1, 100% with ncpus=4 for 4, 75% for 3, and back to no ncpus and 100%. In each test the appropiate number of CPUS + 1 CUDA/CPU<1 were running after the benchmark finished.

Then I let both run, so far so good. It's only been two hours.

I suspended one cpu project with several tasks running, second cpu project had no tasks and a third cpu project only had one task. The caused the single cpu task to run. no other tasks avaialable so it could not run two. Within a few seconds it downloaded a task from project without any and began running it. This is what is suppsoe to happen, and it did, so i'm back to all cpus running.

I checked most of the tabs, all are OK.

No errors logged.

This version seems mostly harmless and more well behaved. I think it is safe for you'll to try.

Again, since it has not been fully tested, I make no guarrantee.

Only bad so far:
Do not try graphics with malaria control app. it crashed both times crashing the malaria task (lost work). graphics was ok with a docking@home task, so maybe it is just the malaria graphics.

I will run more testing later tonight and more extensive testing tomorrow.

---

Alpha Tester ~~ BOINCin since 10-Apr-2004 (2.28) ~~~ Join team USA

Here are the changes:

* client: /proc/N/stat has () around the command name. Remove these.
* client: the preemptability order was wrong
* client: cleaned up CPU scheduler logic somewhat
* client: include precompiled header in rr_sim.cpp so memory leak detection will work.
* MGR: Have the BaseFrame call a function to determine if the selection list should be saved instead of traversing the application pointer. Each view just overrides the function returning a true/false value. We don't have to worry about null pointers and the like.
* MGR: BOINCGUIApp should never need to know how either the views work or the document. Move the code that determines which RPCs should be fired into each of the views. Have the document look for it there.
* MGR: Reduce duplicate code for hiding and showing an application
* MGR: Move some Windows and Mac specific code into functions and streamline the application startup and shutdown rountines.
* MGR: Move the event processing that was in BOINCGUIApp into the BaseFrame.
* MGR: General cleanup.
* MGR: Doxygen comments.
* MGR: Cleanup some warnings
* MGR: Fix show / hide on Mac broken by previous changes
* Rebuild libcurl x86 and x64 (turn off async DNS) for Windows only
* client: revise round-robin simulation to take variable avg_ncpus into account
* Mac SCR: fix some real and potential crash bugs

6.3.21 for windows x86 is running well, I'm beginning some more extensive tests now to see if I can break it.

Computer with 6.3.19 was incorrectly running only 1 CPU + 1 CUDA/CPU<1.

After install to 6.3.21, it immediately ran 2 CPU and 1 CUDA/CPU<1, much better.

---

Here's what you need to do.

If you use <ncpus>#</ncpus> in your cc_config.xml file, remove it. It is no longer needed.

Install new version 6.3.21
boinc_6.3.21_windows_intelx86.exe
boinc_6.3.21_windows_x86_64.exe
boinc_6.3.21_i686-pc-linux-gnu.sh

What you will get is a cpu task for each cpu (ht/dual=2, quad=4) and a GPUGRID task CUDA/CPU<1 running for each gpu. You should always have a gpugrid task running up to max gpus. I only have one, but I'm told it should support multiple gpus now. There will be options in some future version to limit gpus.

Let at least two GPUGRID tasks run and finish. Do not count the one that would be running when you changed versions.

Check the ms/step reported over previous results. Report results in CPUS<1 thread.

If you want you can reduce cpus, so GPUGRID get one cpu core to itself instead of sharing, do this in your project global preferences, either on the website or under advanced settings in boinc manager. Change the use 100% of processors to 50% for HT/DUAL or 75% for Quads. This will run benchmarks and reduce your cpu task usage by one, leaving 1 cpu core for the CUDA/CPU task.

Again let at least two complete GPUGRID tasks run, and check the results. Do not count one if you change in the middle of the run.

It may also be necessary to set 'remove apps from memory' to yes, although now that we have a properly functioning [knock on wood] client that should keep the gpu app running it may not. The GPU app should not get suspended. If you suspend manually, note your gpu elapsed time will be wrong, it will include suspended time, unless app is removed from memory.

If you play games you may want to suspend boinc while playing games. Use the new option in cc_config.xml options, <exclusive_app>gamename.exe</exclusive_app>. Unfortunately this will suspend all boinc while playing, but will save you the trouble of manually suspending and resuming. You probably want to use remove apps from memory also when doing this. Note gamename.exe, without path name, and must be exactly as it will appear in the system, (as in task manager for windows). You can have multiple exclusive_apps in your cc_config.xml file.

The Linux one is again i686, so it wont get WUs from PS3GRID, but crunch those that are already on the system. Why they are not giving out x86-64 compiled clients? Grrr :-(

The Linux i686 will also not d/l any 64 bit work besides the grid so it is totally useless to me :(

Running fine and dandy on Vista32, everything seems good. It appears stable in operation with 2CPU's plus 1GPU, no ncpus settings.....Need to wait for a 64Bit Linux though :(

---

Running fine and dandy on Vista32, everything seems good. It appears stable in operation with 2CPU's plus 1GPU, no ncpus settings.....Need to wait for a 64Bit Linux though :(

Ditto here on Vista 64 bit. 4CPU's + 1GPU. :)

---

Benchmark problem on first run and checking for internet access.

31-Oct-2008 23:46:08 [---] BOINC can't access Internet - check network connection or proxy configuration.
31-Oct-2008 23:46:25 [---] FP benchmark ran only 0.998406 sec; ignoring
31-Oct-2008 23:46:25 [---] FP benchmark ran only 0.982806 sec; ignoring
31-Oct-2008 23:46:25 [---] FP benchmark ran only 0.982806 sec; ignoring

GPU app runs at 17%. With 1% increase, I hope the run time would be less as well.

Edit: seems to reach 18% as well from time to time.

Seems to run okay for me so far on 5 Box's with 5 Wu's running with no setting set for ncpu's, 0%-1% CPU Usage, Win XP Pro 64-Bit all 5 Box's.

PS: Was checking my run times & they seemed to increase across the board from 5.5 Hr's to 7.5 Hr.s running 5 Wu's @ once. I'm going to set the ncpu's to 4 & see what happens ...

Looking good so far. Thanks for the detailed write-up, Keith! Almost worth a FAQ entry, though most of the comments are probably short-lived, as it would only confuse new crunchers to tell them what they don't have to bother about any more ;)

PoorBoy,

if you're getting 0-1% CPU usage something is wrong and you'll see greatly increased crunching times, like you're already saying. With 4+1 tasks my CPU usage is 8 - 13% now, though it did go up to ~20%. And I'd like to draw your attention to Keiths post:

Krunchin-Keith wrote:

If you want you can reduce cpus, so GPUGRID get one cpu core to itself instead of sharing, do this in your project global preferences, either on the website or under advanced settings in boinc manager. Change the use 100% of processors to 50% for HT/DUAL or 75% for Quads. This will run benchmarks and reduce your cpu task usage by one, leaving 1 cpu core for the CUDA/CPU task.

MrS

---

Scanning for our furry friends since Jan 2002

There is some new behaviour: 6.3.21 made a huge increase of my cache size, which seems to be more inline with my settings (1.25 days). Also there was this message:

01/11/2008 12:23:55||[error] Proposed work request 447323.836182 bigger than max 436320.000000

Is the max value project specific (i.e. a server setting) or is it a general BOINC limit? Never mind If it's the former, if it's the latter than 6.3.21 would not obey the BOINC rules :D

MrS

---

Scanning for our furry friends since Jan 2002

There is some new behaviour: 6.3.21 made a huge increase of my cache size, which seems to be more inline with my settings (1.25 days). Also there was this message:

01/11/2008 12:23:55||[error] Proposed work request 447323.836182 bigger than max 436320.000000

Is the max value project specific (i.e. a server setting) or is it a general BOINC limit? Never mind If it's the former, if it's the latter than 6.3.21 would not obey the BOINC rules :D

MrS

I'm getting this message proposed work request bigger than max message as well, but the cache is quickly growing way more than my setting of 1 day. I'm going back to 6.3.19 for now.

PoorBoy, if you're getting 0-1% CPU usage something is wrong and you'll see greatly increased crunching times, like you're already saying. With 4+1 tasks my CPU usage is 8 - 13% now, though it did go up to ~20%. And I'd like to draw your attention to Keiths post:

Krunchin-Keith wrote:

If you want you can reduce cpus, so GPUGRID get one cpu core to itself instead of sharing, do this in your project global preferences, either on the website or under advanced settings in boinc manager. Change the use 100% of processors to 50% for HT/DUAL or 75% for Quads. This will run benchmarks and reduce your cpu task usage by one, leaving 1 cpu core for the CUDA/CPU task.

MrS

I set the Box's to 3 ncpu's & so far they've held to 4 Wu's, 3 Regular CPU Wu's & 1 ... I think I know why the GPU was using 0%-1% only, I'm running that goofy FreeHal Project a little and the runner.exe they use is using up to 25% itself thus denying the GPU any CPU Resource, or so I think thats whats happening.

I'm going to stop running that Project shortly so then things may straighten themselves out a little ... :)

There is some new behaviour: 6.3.21 made a huge increase of my cache size, which seems to be more inline with my settings (1.25 days). Also there was this message:

01/11/2008 12:23:55||[error] Proposed work request 447323.836182 bigger than max 436320.000000

Is the max value project specific (i.e. a server setting) or is it a general BOINC limit? Never mind If it's the former, if it's the latter than 6.3.21 would not obey the BOINC rules :D

MrS

I do not know what this is.

What project is the request for ?

If you don't know,
Try NNW for each project one at a time, for a short while, and see if the message goes away. When it does, turn requests for that project back on, if message resumes that would indicate which project.

PoorBoy, if you're getting 0-1% CPU usage something is wrong and you'll see greatly increased crunching times, like you're already saying. With 4+1 tasks my CPU usage is 8 - 13% now, though it did go up to ~20%. And I'd like to draw your attention to Keiths post:

Krunchin-Keith wrote:

If you want you can reduce cpus, so GPUGRID get one cpu core to itself instead of sharing, do this in your project global preferences, either on the website or under advanced settings in boinc manager. Change the use 100% of processors to 50% for HT/DUAL or 75% for Quads. This will run benchmarks and reduce your cpu task usage by one, leaving 1 cpu core for the CUDA/CPU task.

MrS

I set the Box's to 3 ncpu's & so far they've held to 4 Wu's, 3 Regular CPU Wu's & 1 ... I think I know why the GPU was using 0%-1% only, I'm running that goofy FreeHal Project a little and the runner.exe they use is using up to 25% itself thus denying the GPU any CPU Resource, or so I think thats whats happening.

I'm going to stop running that Project shortly so then things may straighten themselves out a little ... :)

Yes, some other apps do not behave well and hog cpu's.

Can you tell me about the FreeHal app.
Is it a wrapper app ?
Does it use JAVA ?
What system base priority does it run at ?

I have seen similar behavior with malaria control optimizer which is JAVA app inside the wrapper. When it runs, the gpu app sits idle not getting any cpu, accumulating extra elapsed gpu time. Other malaria apps are OK.

I do not know what this is.

What project is the request for ?

If you don't know,
Try NNW for each project one at a time, for a short while, and see if the message goes away. When it does, turn requests for that project back on, if message resumes that would indicate which project.

For me, it was for all 3 projects I participate in (here, CPDN and SAH). I forgot to mention in my previous message that I got a different max value in the error message.

I do not know what this is.

What project is the request for ?

If you don't know,
Try NNW for each project one at a time, for a short while, and see if the message goes away. When it does, turn requests for that project back on, if message resumes that would indicate which project.

For me, it was for all 3 projects I participate in (here, CPDN and SAH). I forgot to mention in my previous message that I got a different max value in the error message.

Different max value for each project ? What values did you get ?

Can you tell me about the FreeHal app.
Is it a wrapper app ?
Does it use JAVA ?
What system base priority does it run at ?

I have seen similar behavior with malaria control optimizer which is JAVA app inside the wrapper. When it runs, the gpu app sits idle not getting any cpu, accumulating extra elapsed gpu time. Other malaria apps are OK.

It uses a Wrapper & a runner.exe *32 that doesn't play nice at all, you have to manually stop the Process when exiting BOINC or it stays running .. Don't know about Jave but there was a lot of talk about it using Pearl ??? The Wrapper & runner both seem to run @ Low Priority though ..

PS: According to the Project the Wu's are supposed to be a non cpu-intensive type or something like that. I informed them already that their full of it ... :)

Different max value for each project ? What values did you get ?

It's the same max value for all projects. I've already rolled back to 6.3.19, so I don't have the exact value anymore, but I think it was 395xxx (I'm sure about it being in the three hundred thousand range, not as certain about the 95xxx).

Different max value for each project ? What values did you get ?

It's the same max value for all projects. I've already rolled back to 6.3.19, so I don't have the exact value anymore, but I think it was 395xxx (I'm sure about it being in the three hundred thousand range, not as certain about the 95xxx).

I cannot reproduce.

What is your connect interval and additional days ?

Are there any other settings that might have changed from deault that might affect this ?

Sorry but without more info it is hard to help.

First off when starting client, check these messages, be sure client is using correct preferences and limits:
11/1/2008 12:21:00 PM||General prefs: from malariacontrol.net (last modified 12-Oct-2008 10:31:26)
11/1/2008 12:21:00 PM||Computer location: home
11/1/2008 12:21:00 PM||General prefs: using separate prefs for home
11/1/2008 12:21:00 PM||Reading preferences override file
11/1/2008 12:21:00 PM||Preferences limit memory usage when active to 1535.04MB
11/1/2008 12:21:00 PM||Preferences limit memory usage when idle to 1842.05MB
11/1/2008 12:21:00 PM||Preferences limit disk usage to 5.28GB

Does the number reported in any way coincide with a memory or disk limit ?