Missing Apps Part 2
log in

Advanced search

Message boards : News : Missing Apps Part 2

1 · 2 · Next
Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19311 - Posted: 2 May 2014, 15:55:49 UTC

If you have a CPU that cannot complete the mini_collatz work unit in less than 8 hours and would like the option to run the micro_collatz work units on your CPU, please let me know and I can enable the CPU apps for the micro_collatz work units.

I disabled the OpenCL for Intel GPUs for large_collatz work units as a majority of the Intel GPUs would not be able to complete the large work units before the deadline.

The deadline for work units has been increased to 14 days, up from 7. It will only take effect on newly created work units. It will not affect the work units that have already been created so there will be a mixture of them for a while. In the mean time, I have increased the grace period to 7 days so those work units with a 7 day deadline can be submitted up to 7 days late and still be awarded credit.

Running longer work units reduces network bandwidth, database size, and the load on the Collatz server. When the solo app was initially released over a year ago, there was a bonus for running the solo work units. It was not removed when the multi-size logic was added. When multi-size logic was moved last week, that bonus remained in effect for all apps including micro, mini, and large apps. The purpose of the bonus was to encourage everyone to run the largest app their machine could run. I don't want to only give the bonus to fast GPUs that can run the large work units since not everyone has a GPU fast enough to run the large work units. So, rather than remove the bonus from micro, mini, and solo work units, the credit bonus will not be awarded when any work unit is completed in less than 30 minutes. That way, it is both our best interests to run longer running work units. That, and the less I have to worry about the server, the more bass fishing I can get in on the weekends.

There are still a number of multi-sized solo work units mixed in with the new and larger solo work units. Please don't abort them unless they will run in under 30 minutes. The sooner they are completed, the sooner everyone will get the larger ones.

Profile Peciak
Avatar
Send message
Joined: 4 Aug 09
Posts: 13
Credit: 669,955,177
RAC: 764
Message 19314 - Posted: 2 May 2014, 19:16:34 UTC

maybe a bonus for:
- assigns more credits if you return within 24 hours
- not delete WU

Robert7NBI
Send message
Joined: 18 Jul 09
Posts: 23
Credit: 1,001,393,853
RAC: 0
Message 19316 - Posted: 2 May 2014, 19:37:52 UTC
Last modified: 2 May 2014, 19:38:00 UTC

I got large (_0 and _1) with deadline 30.5h and now I can't finish older WUs. It isn't fine.

Profile KWSN Checklist
Avatar
Send message
Joined: 25 Jul 09
Posts: 123
Credit: 1,287,549,992
RAC: 0
Message 19318 - Posted: 2 May 2014, 21:19:55 UTC

Thank you for the boundary explanation and reasoning behind it. Where do you plan on hanging the "Gone Fishing" sign for all to see?

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19319 - Posted: 2 May 2014, 22:26:27 UTC - in response to Message 19314.

maybe a bonus for:
- assigns more credits if you return within 24 hours
- not delete WU


Unlike projects like Milkyway which use the output of one result to generate the input for the next work unit, Collatz has no such issues so there is no benefit to having users run shorter work units. In fact the reverse is true as connecting more often requires more network bandwidth, more database queries, more files read/written, etc.

There are valid reasons to abort work units at times, especially since BOINC now counts work from suspended projects against the cache limits. I'd rather have someone abort a work unit if they know they won't finish it than have to wait a couple weeks for it to expire. That's because the server won't have to create a new work unit which increases the database size.

Profile mfc
Send message
Joined: 17 Mar 14
Posts: 5
Credit: 38,211,224
RAC: 0
Message 19321 - Posted: 3 May 2014, 0:40:46 UTC

Dear Slicker:

Some WUs have very short deadline. Please check it.
(However, it is just ok. I can complete the WUs before deadline.)


6288959 5741706 2 May 2014, 21:07:30 UTC 4 May 2014, 3:36:58 UTC In progress --- --- --- Mini Collatz Conjecture v6.00
6288022 5817152 2 May 2014, 20:21:51 UTC 4 May 2014, 2:51:19 UTC In progress --- --- --- Mini Collatz Conjecture v6.00

ps: Issued at 2 May, deadline at 4 May.

Rick A. Sponholz
Avatar
Send message
Joined: 20 Apr 14
Posts: 6
Credit: 481,142,163
RAC: 1,358,351
Message 19322 - Posted: 3 May 2014, 1:39:27 UTC

Slicker,
Thanks for listening to your volunteers, and adjusting the projects accordingly. I've added Large wu's to my accept list, and look forward to letting my gpu's crunch away without hogging your bandwith. Thanks again, Rick
____________

reklov
Send message
Joined: 5 Jan 10
Posts: 3
Credit: 25,141,017
RAC: 0
Message 19325 - Posted: 3 May 2014, 7:29:23 UTC - in response to Message 19311.

It would be nice to disable large wu's also for slower/older NVidea GPUs, like my GT620. They have no chance to finish before deadline but get it sometimes when selecting use other applications if no work is available for your application :-(
____________

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19331 - Posted: 3 May 2014, 14:32:53 UTC - in response to Message 19325.

It would be nice to disable large wu's also for slower/older NVidea GPUs, like my GT620. They have no chance to finish before deadline but get it sometimes when selecting use other applications if no work is available for your application :-(


You can. Edit your collatz preferences and check all apps except large. Likewise, if you only want CUDA or ATI you can select the type of GPU apps as well.

reklov
Send message
Joined: 5 Jan 10
Posts: 3
Credit: 25,141,017
RAC: 0
Message 19333 - Posted: 3 May 2014, 15:31:14 UTC - in response to Message 19331.
Last modified: 3 May 2014, 15:31:58 UTC

It would be nice to disable large wu's also for slower/older NVidea GPUs, like my GT620. They have no chance to finish before deadline but get it sometimes when selecting use other applications if no work is available for your application :-(

You can. Edit your collatz preferences and check all apps except large. Likewise, if you only want CUDA or ATI you can select the type of GPU apps as well.

I know. But I had selected to use Solo and ticked accept other applications if no work is available for the selected ones. And unfortunately got Large. (Solo preferred to reach the next level at wuprop.)
____________

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19340 - Posted: 3 May 2014, 23:29:28 UTC

I made a couple config changes on the server which I really hope will fix the problem of getting WUs you didn't ask for as well as shortening the deadline by several days. The latter seemed to happen if someone else aborted the WU first. There were no settings that should have caused it, but just in case I added a setting telling it not to shorten the deadline.

I also removed a setting which was left over from the multi-size solo WUs which may have been the reason it was sending large instead of solo WUs.

Profile mfc
Send message
Joined: 17 Mar 14
Posts: 5
Credit: 38,211,224
RAC: 0
Message 19355 - Posted: 5 May 2014, 4:44:10 UTC - in response to Message 19311.

... the credit bonus will not be awarded when any work unit is completed in less than 30 minutes.
...


Dear Slicker:

One suggestion on WU assignment:

Stop feed new WU of a specific application type (for example: Mini, 6.04 (opencl_nvidia_gpu)) to the host with last 'Run time' reported < 30 minutes.

This filter should not apply to the Large applications. Because there are no larger WUs available for the host.

wandering willie
Send message
Joined: 18 Aug 10
Posts: 9
Credit: 18,304,587
RAC: 0
Message 19356 - Posted: 5 May 2014, 8:26:28 UTC
Last modified: 5 May 2014, 8:43:39 UTC

large wu 5543623 35:49:00 elapse time 27.67% complete deadline of the 03/05/2014.

Just completed the above WU and validated 05/05/2014 236,964,02 credits.

Hope to get another soon back to solo’s and mini’s at the moment.

EDIT I did have a few problems with BOINC screen freezing after a few minutes or after leaving and returning although GPUZ was saying the unit was still running.

Reboot the unit progress updated

Michael

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19358 - Posted: 5 May 2014, 14:08:20 UTC - in response to Message 19355.

... the credit bonus will not be awarded when any work unit is completed in less than 30 minutes.
...


Dear Slicker:

One suggestion on WU assignment:

Stop feed new WU of a specific application type (for example: Mini, 6.04 (opencl_nvidia_gpu)) to the host with last 'Run time' reported < 30 minutes.

This filter should not apply to the Large applications. Because there are no larger WUs available for the host.


I like the idea, but it is a lot easier said than done. For example, I just learned this past week that when BOINC sends WUs our that were previously aborted, it cuts the deadline in half on the re-send. I had to go and change code that is not expected to be changed by the project admins in order to turn that feature off which means having to make that change every time I upgrade the server software from now on. And people wonder why "upgrades" to the latest server version are avoided.

There are probably about 100 different code files which make up the BOINC scheduler, the part of BOINC responsible for sending out WUs to hosts. Any little change requires the equivalent of Sheldon staring at a white board for hours or even days on end (e.g. from the TV show The Big Bang Theory)trying to decode the logic and figure out how to bypass certain features without breaking others.

For CPUs, I've now figured out where to inject the code to compare the rated integer operations per second (IOPS) for the Micro WUs and could implement that as well for the Mini WUs that are CPU based thereby "forcing" people to run one or the other. Unfortunately, GPUs don't provide IOPS and I haven't seen any benchmark software (e.g. Dhrystone) for GPUs that I could use.

Even if I were to do so, I've also had people complain about getting longer running WUs who want to run short ones even if they get less credit. Or, they have a really fast GPU capable of running the longer WUs but only want short WUs because they have it throttled to 50% due to heat. For exampe, my Toshiba Qosmio laptop which is supposed to be a "gamer" laptop has an nVidia GTX 770M GPU but such lousy cooling that I really can't use it for crunching at all!

So, it is a good suggestion. I just don't know whether it is possible and even if it is, what percentage of people don't want it to work that way.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 19359 - Posted: 5 May 2014, 14:12:51 UTC - in response to Message 19356.

large wu 5543623 35:49:00 elapse time 27.67% complete deadline of the 03/05/2014.

Just completed the above WU and validated 05/05/2014 236,964,02 credits.

Hope to get another soon back to solo’s and mini’s at the moment.

EDIT I did have a few problems with BOINC screen freezing after a few minutes or after leaving and returning although GPUZ was saying the unit was still running.

Reboot the unit progress updated

Michael


If the video response is horrible or the screen "blinks" and then the WU no longer progresses, the config settings are too high. Reduce the kernels_per_reduction by 1 or 2 and that may help. If that isn't enough, reduce the items_per_kernel setting by 1 or 2. If that still isn't enough, reduce the threads by 1 or 2 as well. If that still isn't enough, reduce the kernels_per_reduction by another 1 or 2, etc. until the GPU load hovers around 98 percent. You should have reasonable response at 98% or lower utilizaiton.

Profile mfc
Send message
Joined: 17 Mar 14
Posts: 5
Credit: 38,211,224
RAC: 0
Message 19366 - Posted: 6 May 2014, 9:43:40 UTC - in response to Message 19358.


I like the idea, but it is a lot easier said than done...

For CPUs, I've now figured out where to inject the code to compare the rated integer operations per second (IOPS) for the Micro WUs and could implement that as well for the Mini WUs that are CPU based thereby "forcing" people to run one or the other. Unfortunately, GPUs don't provide IOPS and I haven't seen any benchmark software (e.g. Dhrystone) for GPUs that I could use.

...


Dear Slicker:

Yes, I know it is complicate to make any changes on the system.
Please don't mind for my (crasy) ideas.

About the GPU's IOPS, you may consider estimate it on each WU reported
and update to the host's profile.
(IOPS = Estimated computation size / Elapsed time)

However, a host may have multiple GPUs with variant computation power
and, as you said, someone may just want to run short WUs for any reason.
It is just for your reference.

:)

farnost litomysl
Send message
Joined: 3 Sep 09
Posts: 3
Credit: 3,714,387,388
RAC: 0
Message 19416 - Posted: 10 May 2014, 0:36:54 UTC - in response to Message 19366.

Hallo.
Work from 2th May and returned 10th May is not validate.
Where is your promise to prolong deadline 7+7 days?
See: http://boinc.thesonntags.com/collatz/results.php?hostid=132838&offset=0&show_names=0&state=6&appid=
One week of work lost.

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 19417 - Posted: 10 May 2014, 2:17:47 UTC - in response to Message 19416.

One week of work lost.


They were validated 10 May 2014, 0:07:44 UTC

farnost litomysl
Send message
Joined: 3 Sep 09
Posts: 3
Credit: 3,714,387,388
RAC: 0
Message 19421 - Posted: 10 May 2014, 8:54:01 UTC - in response to Message 19417.

No.
93 large wus were canceled by server, only 2 (by words two) wus were validated. Sorry, It´s unfair.
All wus were sent ready and BM said: "scheduler request completed"
170 hours computing time (3 times HD7970)now lost and 102 kWh energy too.
Thank you.

Profile JayPi
Send message
Joined: 25 Sep 11
Posts: 5
Credit: 1,691,845,339
RAC: 4,094,460
Message 19424 - Posted: 10 May 2014, 11:14:03 UTC

I crunched successfully a large WU on my Intel GPU (i7-3770K) within the old Deadline of 7 days:
http://boinc.thesonntags.com/collatz/workunit.php?wuid=5754787
The CPU is running on normal Speed, not overclocked.
Why are the large WUs now disabled and the Deadline expanded to 14 days? If i dont want the Large WUs, i can disable it.

1 · 2 · Next
Post to thread

Message boards : News : Missing Apps Part 2


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.