Large collatz scedule timing problems
log in

Advanced search

Message boards : Windows : Large collatz scedule timing problems

Author Message
Profile Lonely
Avatar
Send message
Joined: 1 Sep 14
Posts: 3
Credit: 5,523,401
RAC: 0
Message 19832 - Posted: 8 Sep 2014, 9:12:14 UTC

Have had 2 of these increase their original start estimated times. One @ 460 hours and after 196 hours has decreased down to 365 with less than 6 days remaining?

The second unit has shown similar problems and cannot complete in the time remaining... less than 6 days?

Both units have been on contiuous operation since download - 1 / 2 September.

Profile Lonely
Avatar
Send message
Joined: 1 Sep 14
Posts: 3
Credit: 5,523,401
RAC: 0
Message 19833 - Posted: 8 Sep 2014, 10:25:16 UTC

Have bitten the bullet and terminated both units as they will not complete in time = a waste of almost 300 hours. :-(

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3242
Credit: 1,687,119,727
RAC: 6,188,584
Message 19834 - Posted: 8 Sep 2014, 11:27:11 UTC - in response to Message 19833.

Have bitten the bullet and terminated both units as they will not complete in time = a waste of almost 300 hours. :-(


You might want to go into your setting and specify that you only get the smaller units then. You have your pc's hidden so I can't see what your specs are so can't be sure that's the problem.

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: 2
Message 19836 - Posted: 8 Sep 2014, 13:35:19 UTC

Until you have completed a dozen or more WUs for an application, the estimates are terrible. That's because
a) they are based on floating point arithmetic and Collatz uses integer math
b) there are no whetstone or dhrystone benchmarks for GPUs or ASICs that give a true comparison to the FLOPS (floating point operations per second) reported by CPUs.

In addition, some believe:

The current method works tolerably well for all projects. Strict linear based
on % complete will fail miserably for fairly high proportion of projects.


I guess they have a very high tolerance. Way higher than mine anyway.

There's supposed to be a flag that can be set in newer clients that is supposed to assume linear progression of the work done, but I can't seem to find it in the documentation.

Jon Fox
Send message
Joined: 6 Sep 09
Posts: 36
Credit: 351,207,997
RAC: 258,021
Message 19845 - Posted: 9 Sep 2014, 14:49:27 UTC - in response to Message 19836.

Is this the flag you were looking for:

<fraction_done_exact/>


--
jon

Profile Lonely
Avatar
Send message
Joined: 1 Sep 14
Posts: 3
Credit: 5,523,401
RAC: 0
Message 19847 - Posted: 11 Sep 2014, 11:17:19 UTC

@ Mikey - As a newbie these were include in the first lot of downloads. Every thing looked OK until they started to add time onto the reduction times.

Have since checked the way to limit type of units and all seems to be falling back into place.

One unit was on a 2.4 dual proc 8gb ram and an nVidia wth 512 ram. This one I can understand?

The 2 others are i7 2.4 quads with 8gb ram + nVidia and Intel 4000 GPU's with 2048 and 1024 ram respectively.

My thanks to all for your replies. :thumbs-up:

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: 2
Message 19848 - Posted: 11 Sep 2014, 13:09:39 UTC - in response to Message 19845.

Is this the flag you were looking for:

<fraction_done_exact/>


--
jon



Thanks. That's the one!Per the documentation it goes in the app_config.xml file.

Jon Fox
Send message
Joined: 6 Sep 09
Posts: 36
Credit: 351,207,997
RAC: 258,021
Message 19849 - Posted: 11 Sep 2014, 19:12:12 UTC - in response to Message 19848.
Last modified: 11 Sep 2014, 19:18:12 UTC

Glad to help. Yes, app_config.xml it is...here's mine:

<app_config>
<app>
<name>solo_collatz</name>
<max_concurrent>4</max_concurrent>
<fraction_done_exact/>
<gpu_versions>
<gpu_usage>0.75</gpu_usage>
<cpu_usage>1.00</cpu_usage>
</gpu_versions>
</app>
<app>
<name>mini_collatz</name>
<max_concurrent>4</max_concurrent>
<fraction_done_exact/>
<gpu_versions>
<gpu_usage>0.75</gpu_usage>
<cpu_usage>1.00</cpu_usage>
</gpu_versions>
</app>
<name>large_collatz</name>
<max_concurrent>4</max_concurrent>
<fraction_done_exact/>
<gpu_versions>
<gpu_usage>1.00</gpu_usage>
<cpu_usage>1.00</cpu_usage>
</gpu_versions>
</app>
</app_config>

--
jon

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: 2
Message 19851 - Posted: 12 Sep 2014, 2:57:08 UTC

I also added the fraction_done_exact code to the server so it __should__ be automatically set for those running the BOINC 7.3.x or later client without the use of an app_config.xml file.


Post to thread

Message boards : Windows : Large collatz scedule timing problems


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.