Phasing out Multi-Size
log in

Advanced search

Message boards : News : Phasing out Multi-Size

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: 2491
Credit: 702,712,176
RAC: 57,774
Message 19121 - Posted: 23 Apr 2014, 23:25:15 UTC

Due to a variety of server scheduling problems that have not been able to be resolved, Collatz will be phasing out the multi-size workunits over the next week or two and going back to using multiple applications. You will once again control your own destiny and be able to choose the size WUs by the application chosen (micro, mini, standard, and large). The big addition (pun intended) that will be the first to be deployed is the large_collatz application which will be 32 times the size of the largest multi-size workunit. It is aimed at the fastest GPUs to reduce the upload/download and workunit startup overhead of running smaller workunits. That should result in a higher RAC as well as a reduced load on the Collatz server.

Profile Phil1966
Send message
Joined: 25 Dec 11
Posts: 17
Credit: 175,424,030
RAC: 887,452
Message 19128 - Posted: 24 Apr 2014, 4:34:49 UTC - in response to Message 19121.
Last modified: 24 Apr 2014, 4:48:35 UTC

Dear Slicker,

Thank you for the update.

It's a very good idea ! :)

Do you know when the large_collatz will be released, so we can update the .config &/or app_config files accordingly ?

Thank You

Kind Regards

Phil1966

Profile Phil1966
Send message
Joined: 25 Dec 11
Posts: 17
Credit: 175,424,030
RAC: 887,452
Message 19131 - Posted: 24 Apr 2014, 5:57:12 UTC

Sorry, read to fast. Re-Reading and noticed : "over the next week or two" :)

Thank You

Phil1966

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2491
Credit: 702,712,176
RAC: 57,774
Message 19139 - Posted: 24 Apr 2014, 19:11:19 UTC

Update: The micro, mini, and large work generators appear to be working now. Until WUs have completed, I won't know if the validators or assimilators are working properly. At present, these are considered beta apps so if they work, great. If not, I'm not spending time fixing lost credits. The micro WUs take about 3 hours in my cell phone so I should know later today if the server daemons for them are working OK. The large WUs may take 18 hours on a nVidia GTX 570 so I won't know about them until tomorrow. Once I'm sure the daemons are working OK, I'll add the remaining apps where appropriate.

[AF>FAH-Addict.net]toTOW
Send message
Joined: 15 Oct 10
Posts: 15
Credit: 100,193,374
RAC: 138,025
Message 19142 - Posted: 24 Apr 2014, 21:03:41 UTC
Last modified: 24 Apr 2014, 21:16:02 UTC

Can you post a relation between the size of each WU ?

Mini Collatz Conjecture = ?
Solo Collatz Conjecture = ?
Large Collatz Conjecture = 32 * Solo ?
Micro Collatz Conjecture = ?

edit: I enabled only Large WUs on my profile (and I'm allowing to receive work from other app if none were available), but I'm getting :

24/04/2014 23:06:49 | Collatz Conjecture | Sending scheduler request: To fetch work.
24/04/2014 23:06:49 | Collatz Conjecture | Requesting new tasks for ATI
24/04/2014 23:06:51 | Collatz Conjecture | Scheduler request completed: got 0 new tasks
24/04/2014 23:06:51 | Collatz Conjecture | Project has no tasks available


GPU is a R9 290X.

edit again: I got work, but that's only Solo WUs, not Large ones ... :(

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2491
Credit: 702,712,176
RAC: 57,774
Message 19143 - Posted: 24 Apr 2014, 21:58:46 UTC

micro = 6,442,450,944 numbers
mini = 103,079,215,104 numbers (16 times larger than micro)
solo = 1,649,267,441,664 numbers (16 times larger than mini)
large = 26,388,279,066,624 numbers (16 times larger than solo)

Note: Solo will be doubling in size but is currently 824,633,720,832 numbers. Once the other three are tested, I will turn off multi-size on the solo WUs and turn on the new work generator.

As I stated earlier, these are beta apps. Beta = test. So, you won't get any at present unless you have the "Run test applications" selected in your preferences. There were validation issues with several of the apps causing 0 credit even though valid. I will NOT be going back and fixing credit for beta/test apps, so there is some risk involved at this point even if it finishes successfully.

[AF>FAH-Addict.net]toTOW
Send message
Joined: 15 Oct 10
Posts: 15
Credit: 100,193,374
RAC: 138,025
Message 19145 - Posted: 24 Apr 2014, 22:34:18 UTC

Thanks for posting the size ... so if I extrapolate from my current speeds (7.5 minutes for the biggest solo with multi size) :

New solo = 15 minutes
Large = 240 minutes (4 hours)

I have the "Run test applications" options selected in my profile, but I'm still not getting the new WUS ... maybe the server is waiting until my current Solo WU cache is depleted ?

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2491
Credit: 702,712,176
RAC: 57,774
Message 19146 - Posted: 24 Apr 2014, 22:40:34 UTC - in response to Message 19145.

Thanks for posting the size ... so if I extrapolate from my current speeds (7.5 minutes for the biggest solo with multi size) :

New solo = 15 minutes
Large = 240 minutes (4 hours)

I have the "Run test applications" options selected in my profile, but I'm still not getting the new WUS ... maybe the server is waiting until my current Solo WU cache is depleted ?


Correct.

The first thing I do after installing boinc is to enable sched_op_debug in the cc_config.xml file so I can see if the client is really asking for work or not.

[AF>FAH-Addict.net]toTOW
Send message
Joined: 15 Oct 10
Posts: 15
Credit: 100,193,374
RAC: 138,025
Message 19151 - Posted: 25 Apr 2014, 0:49:54 UTC

It is asking for work :

25/04/2014 02:48:23 | Collatz Conjecture | update requested by user
25/04/2014 02:48:26 | Collatz Conjecture | sched RPC pending: Requested by user
25/04/2014 02:48:26 | Collatz Conjecture | [sched_op] Starting scheduler request
25/04/2014 02:48:26 | Collatz Conjecture | Sending scheduler request: Requested by user.
25/04/2014 02:48:26 | Collatz Conjecture | Requesting new tasks for ATI
25/04/2014 02:48:26 | Collatz Conjecture | [sched_op] CPU work request: 0.00 seconds; 0.00 devices
25/04/2014 02:48:26 | Collatz Conjecture | [sched_op] ATI work request: 1193382.70 seconds; 0.00 devices
25/04/2014 02:48:28 | Collatz Conjecture | Scheduler request completed: got 0 new tasks
25/04/2014 02:48:28 | Collatz Conjecture | [sched_op] Server version 703
25/04/2014 02:48:28 | Collatz Conjecture | Project has no tasks available
25/04/2014 02:48:28 | Collatz Conjecture | Project requested delay of 61 seconds
25/04/2014 02:48:28 | Collatz Conjecture | [sched_op] Deferring communication for 00:01:01
25/04/2014 02:48:28 | Collatz Conjecture | [sched_op] Reason: requested by project


But the server isn't sending anything :(

Profile Pooh Bear 27
Avatar
Send message
Joined: 1 Aug 10
Posts: 54
Credit: 108,227,920
RAC: 0
Message 19159 - Posted: 25 Apr 2014, 9:53:16 UTC

So, I am forced to only do micro on ARM? I was doing 3 different sizes before, I think at least mini should also be allowed.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2491
Credit: 702,712,176
RAC: 57,774
Message 19161 - Posted: 25 Apr 2014, 14:55:59 UTC - in response to Message 19159.

So, I am forced to only do micro on ARM? I was doing 3 different sizes before, I think at least mini should also be allowed.


Q: How many Collatz volunteers does it take to change a light bulb?
A: 20. 1 to change it and 19 to say they liked the old burned out one better. ;-)

How long does it take to crunch the various sized WUs on your android device?

Profile JayPi
Send message
Joined: 25 Sep 11
Posts: 5
Credit: 1,064,580,549
RAC: 4,966,102
Message 19162 - Posted: 25 Apr 2014, 15:48:13 UTC

I got my first large WU for my Intel-GPU. 19 minutes for 0.2% ==> it will be running about 6,6 days with a Deadline of 7 days. I stop it, because the deadline give not enough time to safely complete this unit.

Profile Pooh Bear 27
Avatar
Send message
Joined: 1 Aug 10
Posts: 54
Credit: 108,227,920
RAC: 0
Message 19163 - Posted: 25 Apr 2014, 18:39:05 UTC - in response to Message 19161.
Last modified: 25 Apr 2014, 18:46:58 UTC

How long does it take to crunch the various sized WUs on your android device?

Problem is you purge too quickly. It wasn't long for any of the 3 sizes. I think I remember the longer ones being 4-5 hours. The short ones were less than an hour.

Grabbed a micro, it's slated to finish in just under an hour.

gomeyer
Send message
Joined: 1 Mar 12
Posts: 4
Credit: 62,112,081
RAC: 0
Message 19165 - Posted: 25 Apr 2014, 19:43:03 UTC
Last modified: 25 Apr 2014, 19:45:59 UTC

I'm asking for mini and micro for my wimpy GPUs, server sez nothing available.

edit: CUDA only

[AF>FAH-Addict.net]toTOW
Send message
Joined: 15 Oct 10
Posts: 15
Credit: 100,193,374
RAC: 138,025
Message 19166 - Posted: 25 Apr 2014, 20:18:45 UTC

Still no Large WUs for my R9 290X :'(

Profile Gibson Praise
Send message
Joined: 27 Feb 14
Posts: 4
Credit: 331,588,779
RAC: 934,816
Message 19168 - Posted: 25 Apr 2014, 21:13:49 UTC

just received a large WU. expected time on r270: 1446 hours and change.
I don't think the scheduler got that right!

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 19169 - Posted: 25 Apr 2014, 21:23:42 UTC - in response to Message 19168.

I don't think the scheduler got that right!


Run it - it will settle. Its the usual BOINC "learning curve" when it runs a new type of app, it goes back square one with estimated run time on first run.

It has no affect at all on the app, watch the first figure, "time taken" figure, and extrapolate to 100% after its done 10% or so.

Its a BOINC "quirk" of many many years duration, nothing Collatz can do about it.

Profile Tarmo Ilves
Send message
Joined: 6 Jun 10
Posts: 6
Credit: 408,876,839
RAC: 1,531,426
Message 19170 - Posted: 25 Apr 2014, 21:29:49 UTC

After 3h 27min 53,6% completed.
Large Collatz on HD 7990

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 19172 - Posted: 25 Apr 2014, 22:03:51 UTC - in response to Message 19170.
Last modified: 25 Apr 2014, 22:05:31 UTC

After 3h 27min 53,6% completed.


Double check the file: large_collatz_6.04_windows_x86_64__opencl_amd_gpu.CONFIG

Suspect it may be blank because that's on the slow side for a 7990, you should be either side of 4.5hrs+ for the complete file depending on GPU clocks that you have set.

Mine (on 7970s - clocks 1185/1685) is below:

verbose=1
items_per_kernel=22
kernels_per_reduction=9
threads=8
sleep=1

Depending on how it runs on the 7990, you might need to reduce "items per kernal" to 21

chf1949
Send message
Joined: 18 Mar 11
Posts: 16
Credit: 1,363,723,776
RAC: 0
Message 19173 - Posted: 25 Apr 2014, 22:04:32 UTC - in response to Message 19166.
Last modified: 25 Apr 2014, 22:34:44 UTC

Still no Large WUs for my R9 290X :'(


In BOINC Manager, make sure your local preferences are set to accept , say, 2 days of work... If ur local preferences queue is small, the 'system' will not send anything because it thinks there will not be enough time to run the large wu's.

Also in Collatz preferences make sure you have the 'run test units' box ticked.

Plus, in the 'run only selected application' box, expand it and make sure 'large' work units is selected.

Then restart Boinc so it reads the new preferences. After much fiddling, that is what worked for me.

1 · 2 · Next
Post to thread

Message boards : News : Phasing out Multi-Size


Main page · Your account · Message boards


Copyright © 2016 Jon Sonntag; All rights reserved.