Thread 'My Wish List'

Message boards : BOINC client : My Wish List
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 8 · Next

AuthorMessage
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 36893 - Posted: 17 Feb 2011, 12:51:33 UTC
Last modified: 22 Feb 2011, 15:20:36 UTC

A new beginning, a new thread. The other one was a bit filled up, even though only at ~150 posts.

For clarity, please do not add your signature when you post to this thread.

Before you add to this thread, see if your request isn't already in the list of past requests and what answer the developers gave them.

Seeing how just about everyone wants to start their own "wish" thread these days in the Client forum, I am moving this thread from the Manager forum to the Client forum. Not that it'll attract much more attention here, but hey I can try. ;-)
ID: 36893 · Report as offensive
Fred

Send message
Joined: 18 Feb 11
Posts: 4
United States
Message 36903 - Posted: 18 Feb 2011, 14:25:28 UTC

Here's a few things I'd like to see. I didn't see them in your FAQ reject list (which was quite useful). By the way, is there a similar list of things already under development?

1. I recently installed a Nvidia card with Cuda capability, and I've had a lot of cuda tasks fail from the 3 projects I run. The error messages from each project were different, but I have now concluded that the cause was the same - my daughter switched users in WinXP while a cuda task was running, and it errored out. Other tasks in the cache then sequentially started and failed after 2 or 3 seconds each until all tasks on hand have failed, or the other user returns to my ID. One project has even built in a 10 minute delay to minimize this problem.

One of the helpful folks at Einstein confirmed that this is a known issue, and it is blamed on how Windows handles multiple users. I confirmed with a test Einstein task under Win XP SP3. I understand these errors also occur with Win 7, according to another user requesting help at Einstein. The tasks don't fail if you manually suspend gpu processing before switching users. I taught my daughter and grandson how to snooze the GPU before switching to their id's, but they'll forget from time to time. They can log me off, but that will stop cpu tasks. If is possible to detect that Windows is switching users and suspend GPU processing, I wish BOINC would do that for me. If BOINC can't do it, the projects ought to look and see if they can improve their error handling to suspend the task rather than exit when they lose contact with the GPU. Realize that BOINC can't do anything about how project applications handle errors, but maybe you can help the projects if the errors can be avoided.

2. Would like to see the number of days' work to keep on hand become a project preference rather than a global one. My problem with one size fits all is that SETI has had a lot of outages lately and gives generous deadlines, so I want a lot of work on hand. But if I set the preference for that, certain unnamed projects will bombard me with too many short deadline tasks. Surely they aren't trying to get more than their allocated share of resources....

3, Would like to be able to change resource shares from BOINC Manager and have that info sent to the project servers. Currently have to visit each site to do this. Just a nuisance item and I realize it would require changes at both ends. Maybe this one is for a major redo someday. I realize the account managers will do this for me, but that's about all they have to offer me, and I don't like some of their features.

4. Saw in your FAQ that a future BOINC Mgr. release may allow editing cc_config.xml from BOINC Mgr. That will help. I've run BOINC for years, but I only recently found the information about that file by running searches on this site. I'm not sure 2.1 mil users all over the world will do that research to understand this file, so I wonder if a few of the most commonly used items could be preferences or menu options in BOINC Manager. Some to consider:

a. Use or not use a specific GPU. This one is why i did the research to find out about the config file. BOINC kept pinging SETI for GPU tasks for my motherboard's ATI GPU. Other sites have a project preference to not use the ATI GPU, but SETI doesn't because they don't yet support the ATI GPU's. Only way to stop the task requests was the config file. This subject of using one, all, or no gpu's is all over the help desk forums - could it be handled in BOINC Manager preferences or options? Something on the menus is morelikely to be found.

b. Zero out the long term debt- useful when project balances get out of kilter and I want a fresh start.

Don't know that you need it, but running latest BOINC Mgr, Win XP is up to date and using latest NVIDIA drivers.

ID: 36903 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 36904 - Posted: 18 Feb 2011, 14:53:17 UTC - in response to Message 36903.  

By the way, is there a similar list of things already under development?

Of sorts, in a way: http://boinc.berkeley.edu/trac/wiki/DevProjects

1. If is possible to detect that Windows is switching users and suspend GPU processing, I wish BOINC would do that for me.

It's on the TODO list, and only this morning I emailed the developers on this again. Probably after reading your thread. ;-)

2. Would like to see the number of days' work to keep on hand become a project preference rather than a global one.

Not gonna be done, as far as I know. It'll screw up the CPU/GPU schedulers in BOINC to no end and we're just about now seeing the light at the end of the scheduling tunnel, after they've been changed and redone for the past 3 years.

3, Would like to be able to change resource shares from BOINC Manager and have that info sent to the project servers. Currently have to visit each site to do this.

Not going to be done. It's something of a security risk as well, to have your computer dictate to a project server what it should do or set. As you said already, you can do this through an account manager. So either.. or..

4. This subject of using one, all, or no gpu's is all over the help desk forums - could it be handled in BOINC Manager preferences or options?

b. Zero out the long term debt- useful when project balances get out of kilter and I want a fresh start.

These are something that the developers deem 'too technical' for the beginner and average user to use. So neither's going to be a menu item, on its own. Even editing the cc_config.xml is something for advanced users only.

Oh, PS: "useful when project balances get out of kilter" .. and you want to set how much work you can cache per project?? How quickly do you think your project balances will be out the window doing that? ;-)
ID: 36904 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 36906 - Posted: 18 Feb 2011, 15:45:20 UTC - in response to Message 36904.  

1. If is possible to detect that Windows is switching users and suspend GPU processing, I wish BOINC would do that for me.

It's on the TODO list, and only this morning I emailed the developers on this again. Probably after reading your thread. ;-)

I must say, I just tested this for the developers and found:
1. BOINC continues to run under the other user account.
2. Both CPU and GPU will continue unaffected.
3. No crashes, but no suspension either.

And that's on Windows 7 - 64bit with BOINC 6.12.15. So you could test it later today when it's going to be released for testing.
ID: 36906 · Report as offensive
Fred

Send message
Joined: 18 Feb 11
Posts: 4
United States
Message 36909 - Posted: 18 Feb 2011, 18:31:03 UTC - in response to Message 36906.  

Will give 6.12.15 a test drive with XP SP3 and let you know, probably tomorrow. I also saw your comments on Einstein's forum where we've discussed this issue. Thanks for your interest in this problem. I saw the list of development items that you mentioned, and much of it is just too tech for us ordinary mortals. I did see a few items of interest, hope they get done.

The other items in my list are just nuisance items and I have no problem with your response. I know the workarounds for all except the days on hand. Will just keep that number low and hope for happy uptimes at the projects.

Fred
ID: 36909 · Report as offensive
Fred

Send message
Joined: 18 Feb 11
Posts: 4
United States
Message 36931 - Posted: 19 Feb 2011, 23:26:29 UTC

Well, tried 6.12.15 under Win XP SP3 and results are wierd. Cuda tasks seemed to continue running when I switched users, then suspended when I returned to my ID - they suspended for about 2 minutes (my preference setting), then the Einstein task resumed from last checkpoint although my pref is set to keep in memory and my activity tab setting was to run always. Changed preference to run always, but it did the same thing - suspended on return for about 2 minutes.

The Seti task was okay the first time I switched back, but on the second test it suspended for 10-20 minutes, and then errored out wih exit code 1 (0x1)- Incorrect function.

Tried the Einstein task a couple of more times - it eventually errored out exit code 1020 (0c3fc with the message 'Cannot create a symbolic link in a registry key that already has subkeys or values'. That's the same one we talked about on Einstein's forum - the error we're getting with 6.10.58 from switching users.

Also noted this test version started a popup box on the other users when I logged on, offering to view a client or something like that. Maybe that's just for testing, don't think you want it in production.

You'd better turn this over to experienced testers to try it - make sure they stay on other id for more than 2 minutes, etc.

Have preserved the activity log to send if you want it - won't post it here for benefit of other readers. Also have one of the stderr.txt files. Going back to 6.10.58 now.

Fred
ID: 36931 · Report as offensive
Fred

Send message
Joined: 18 Feb 11
Posts: 4
United States
Message 36932 - Posted: 20 Feb 2011, 2:41:36 UTC

Caution, those results in my last post may be tainted. My computer bluescreened half an hour after reinstalling 6.10.58 - graphics driver in loop. That's first one of those with this card (3 weeks). Hope it's because I didn't reboot after changing BOINC Mgr. versions or something like that, but could have been related to those tasks hanging up and the errors.

Card is EVGA GeForce 240 GT, driver is +.14.12.6658, cuda version 3020, 1G mem.

BTW, when I suggested staying on the other id for 2 min, I meant 2 inactive minutes in case task suspends and resumes when inactive more than prefs setting. - think that's why some of my tasks failed and others didn't.

Fred
ID: 36932 · Report as offensive
Pepo
Avatar

Send message
Joined: 3 Apr 06
Posts: 547
Slovakia
Message 36965 - Posted: 22 Feb 2011, 15:25:52 UTC - in response to Message 36906.  

1. If is possible to detect that Windows is switching users and suspend GPU processing, I wish BOINC would do that for me.

It's on the TODO list, and only this morning I emailed the developers on this again. Probably after reading your thread. ;-)

I must say, I just tested this for the developers [...]
And that's on Windows 7 - 64bit with BOINC 6.12.15. So you could test it later today when it's going to be released for testing.

Note: changeset [trac]changeset:23068[/trac] has been committed after tagging the 6.12.15 release.

Peter
ID: 36965 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 36966 - Posted: 22 Feb 2011, 15:35:25 UTC - in response to Message 36965.  

Yes Peter, as a consequence of me testing that with 6.12.15. We wanted to know if it was broken, or not. Damn, it was broken.
ID: 36966 · Report as offensive
Liuqyn

Send message
Joined: 21 Sep 09
Posts: 5
United States
Message 36988 - Posted: 23 Feb 2011, 11:34:56 UTC

the only thing currently on my wish list is, not having boinc assign new host id's so often... I like to track stats for each host across all projects but when it generates new id's it makes it impossible without keeping EVERY project attached all the time. also some projects don't allow host merging so I end up with multiple hosts for the same computer.
ID: 36988 · Report as offensive
Jesse Viviano

Send message
Joined: 14 Feb 11
Posts: 63
United States
Message 37004 - Posted: 25 Feb 2011, 3:36:42 UTC
Last modified: 25 Feb 2011, 3:37:03 UTC

I have two wishes:

  • Make the scheduler hyper-threading aware. I have posted this in another thread.
  • Add DirectCompute to the possible targets for Windows BOINC clients. The reasoning is as follows: Intel is unfortunately the market leader for GPUs. Currently, the only way to use them in GPGPU applications is to use DirectCompute because Intel's graphics chipsets do not support OpenCL. One possible drawback is that this will slow CPU tasks because it will compete with the CPU cores for memory throughput. One additional positive is that programmers do not need to compile separate paths for ATI/AMD and nVidia like you need to do with OpenCL.

ID: 37004 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 37008 - Posted: 25 Feb 2011, 8:00:12 UTC - in response to Message 37004.  

*Add DirectCompute to the possible targets for Windows BOINC clients. The reasoning is as follows: Intel is unfortunately the market leader for GPUs. Currently, the only way to use them in GPGPU applications is to use DirectCompute because Intel's graphics chipsets do not support OpenCL.

Not sure where you got this information from, but the Intel HD 2000 and 3000 graphics chips will both support OpenCL, just as the CPUs themselves will support it. I believe you were set on the wrong foot by looking into the release notes on the present Intel OpenCL Alpha SDK, which says it's for the CPU only. This doesn't mean that in the future their graphics won't be supported.

One possible drawback is that this will slow CPU tasks because it will compete with the CPU cores for memory throughput. One additional positive is that programmers do not need to compile separate paths for ATI/AMD and nVidia like you need to do with OpenCL.

Another (bigger) drawback is that this form of computing is only supported on Windows Vista and 7, not on 2000 and XP, which are still used on most Windows computers out there.

Thus you're breaking BOINC its cross-platform compatibility, where one can compile the same BOINC version from the same source code, be it for Linux, Mac or Windows, without needing extra programs, APIs or code.
ID: 37008 · Report as offensive
ProfileJord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15542
Netherlands
Message 37015 - Posted: 25 Feb 2011, 22:17:04 UTC - in response to Message 37004.  

Now, before you scream bloody murder at me, I did send both of them to the developers, and got the next answer back.
Rom Walton wrote:
IIRC, hyper-threading only added a 5-10% overhead to processing a task, both tasks spend more time waiting for memory fetches than fighting over the FPU.

As far as DirectCompute is concerned, you are correct in that it is a Windows only technology.

Intel will at some point provide a OpenCL driver for their videochips.

Eventually all the vendors implementations will mature to the point that a project will be able to release one binary and it'll compile and optimize for the host machine at runtime.

----- Rom
ID: 37015 · Report as offensive
Claggy

Send message
Joined: 23 Apr 07
Posts: 1112
United Kingdom
Message 37017 - Posted: 25 Feb 2011, 22:45:28 UTC

On my wish list is per app DCF for 6.12.x,

i've been running a third party version of 6.10.58 with per app dcf for 6 months now, it just works, somewhat better than Seti's and Collatz's server's scaling the <rsc_fpops_est> per app, no knee jerk reaction after 10 validations!!,

Claggy
ID: 37017 · Report as offensive
Jesse Viviano

Send message
Joined: 14 Feb 11
Posts: 63
United States
Message 37057 - Posted: 2 Mar 2011, 15:12:08 UTC
Last modified: 2 Mar 2011, 15:13:08 UTC

Here is one idea to make BOINC more intelligent in its decision on when to contact the scheduler. In order to clarify things, I will define an execution resource as an entity that can run one task. For example, a logical CPU core or a GPU count as a resource.

Make the client hold a scheduler request until one of these conditions have been met:

  • All files to upload to that project the scheduler request is being generated for have been successfully uploaded.
  • An attempt to upload a file to the project in which the scheduler request is being generated for has failed.
  • One or more execution resources are sitting idle, and the scheduler request is requesting work for that execution resource.
  • There are no more work units that can keep an execution resource busy after one of its tasks finish. For example, a user has 12 logical cores, 12 single-threaded tasks, and one 12-thread task. If one of the single-thread tasks finishes, one logical CPU core will be idle. Therefore, a request for more CPU tasks must be made immediately instead of waiting for all uploads to complete because there will be 11 single-threaded tasks executing until the work scheduler decides to swap in the 12-thread task.


This will cut down on the amount of work units that have been uploaded but not reported yet, but will still keep work throughput at a maximum.

ID: 37057 · Report as offensive
Pepo
Avatar

Send message
Joined: 3 Apr 06
Posts: 547
Slovakia
Message 37058 - Posted: 2 Mar 2011, 15:53:22 UTC - in response to Message 37057.  

Principially yes, just one detail:

Make the client hold a scheduler request until one of these conditions have been met:

  • One or more execution resources are sitting idle, and the scheduler request is requesting work for that execution resource.
  • There are no more work units that can keep an execution resource busy after one of its tasks finish. For example, a user has 12 logical cores, 12 single-threaded tasks, and one 12-thread task. If one of the single-thread tasks finishes, one logical CPU core will be idle. Therefore, a request for more CPU tasks must be made immediately instead of waiting for all uploads to complete because there will be 11 single-threaded tasks executing until the work scheduler decides to swap in the 12-thread task.

This will cut down on the amount of work units that have been uploaded but not reported yet, but will still keep work throughput at a maximum.


The client must prevent any execution resource potentially becoming idle and has to try to keep at least "work_buf_****_days" of work per resource available in advance. It is already late if one resource already became idle.

Peter
ID: 37058 · Report as offensive
Jesse Viviano

Send message
Joined: 14 Feb 11
Posts: 63
United States
Message 37062 - Posted: 2 Mar 2011, 17:01:11 UTC - in response to Message 37058.  

Principially yes, just one detail:

Make the client hold a scheduler request until one of these conditions have been met:

  • One or more execution resources are sitting idle, and the scheduler request is requesting work for that execution resource.
  • There are no more work units that can keep an execution resource busy after one of its tasks finish. For example, a user has 12 logical cores, 12 single-threaded tasks, and one 12-thread task. If one of the single-thread tasks finishes, one logical CPU core will be idle. Therefore, a request for more CPU tasks must be made immediately instead of waiting for all uploads to complete because there will be 11 single-threaded tasks executing until the work scheduler decides to swap in the 12-thread task.

This will cut down on the amount of work units that have been uploaded but not reported yet, but will still keep work throughput at a maximum.


The client must prevent any execution resource potentially becoming idle and has to try to keep at least "work_buf_****_days" of work per resource available in advance. It is already late if one resource already became idle.

Peter

I created the third and fourth conditions because there are times when someone is not attached to enough projects to keep all of the execution resources busy. For example, if an ATI/AMD GPU user is attached to MilkyWay@home as his or her only GPU-capable project and that project's server suffers problems like it is doing now rather often, these conditions will ensure that requests are made immediately instead of waiting on any uploads.
ID: 37062 · Report as offensive
skildude

Send message
Joined: 24 Mar 09
Posts: 12
United States
Message 37157 - Posted: 12 Mar 2011, 16:58:04 UTC

allow columns on the task page of the BOINC manager to be set by client. in other words I can arrange my columns the way I want to see them.
ID: 37157 · Report as offensive
skildude

Send message
Joined: 24 Mar 09
Posts: 12
United States
Message 37159 - Posted: 12 Mar 2011, 19:52:09 UTC - in response to Message 37157.  
Last modified: 12 Mar 2011, 19:52:16 UTC

not just the columns in the task portion but the tabs as well. we should be able to slide the tabs left or right to our preferences
ID: 37159 · Report as offensive
Wilgard

Send message
Joined: 8 Jul 10
Posts: 10
France
Message 37251 - Posted: 21 Mar 2011, 18:18:56 UTC

Hello,

I would like to specify two rates of a time processor.
Actually you can specify the time table for processor using but we can specify only one rate of time processor.

During the day I use the rate to 40% to avoid the fan to run too fast (on my work laptop) but every evening I have to change this rate to 100% during the night and every morning I remodify to 40% for the day.

Do you have a solution ? or could you add this feature to the next release ?

Many thanks.

Wilgard
ID: 37251 · Report as offensive
1 · 2 · 3 · 4 . . . 8 · Next

Message boards : BOINC client : My Wish List

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.