Only half (50%) elapsed time registered

log in

Advanced search

Message boards : Number crunching : Only half (50%) elapsed time registered

1 · 2 · 3 · Next
Author Message
Profile JohnMD
 
Avatar
Send message
Joined: 11 Dec 10
Posts: 29
Credit: 199,247
RAC: 111
Total hours: 303,495
Message 5091 - Posted: 17 May 2017, 10:52:00 UTC

For each 6-hour execution on my N2920, only 3 hours is being counted.

Sergey Kovalchuk
     
Send message
Joined: 27 Oct 15
Posts: 111
Credit: 381,180
RAC: 260
Total hours: 1,261,137
Message 5092 - Posted: 17 May 2017, 11:06:04 UTC - in response to Message 5091.

Check error log - "facteur correction: 0.500000 (4/8.010000)"

You ran 8 threads on 4 cores
Probably you have problems with monkeys, which under Windows are considered to be full CPU applications
Use the app_config to decrease the usage to 0.001 CPU

Profile JohnMD
 
Avatar
Send message
Joined: 11 Dec 10
Posts: 29
Credit: 199,247
RAC: 111
Total hours: 303,495
Message 5094 - Posted: 17 May 2017, 14:01:19 UTC

Thanks Sergey.
I suppose it's stupid to ask why monkeys isn't tagged NCI - and why WUProp doesn't register actual runtime as supplied by the operating system !
I've applied your fix and look forward to more realistic stats :)

Profile Steve Hawker*
       
Send message
Joined: 31 Oct 12
Posts: 125
Credit: 642,670
RAC: 117
Total hours: 1,555,843
Message 5099 - Posted: 17 May 2017, 22:04:18 UTC - in response to Message 5094.

Thanks Sergey.
I suppose it's stupid to ask why monkeys isn't tagged NCI - and why WUProp doesn't register actual runtime as supplied by the operating system !
I've applied your fix and look forward to more realistic stats :)


The whole Monkeys thing is a mess. I have no idea why the previous hours are not added to the current versions. They're the same app, all that's changed is the project name. When CERN consolidated their apps, for the most part things were swapped here without too much trouble.

As for the NCI vs CPU debacle, this is a continuation of the problems Goofy was having on the previous server.

But after all this is BOINC and YMMV is the rule.

Sergey Kovalchuk
     
Send message
Joined: 27 Oct 15
Posts: 111
Credit: 381,180
RAC: 260
Total hours: 1,261,137
Message 5103 - Posted: 18 May 2017, 12:52:31 UTC - in response to Message 5094.

Thanks Sergey.
I suppose it's stupid to ask why monkeys isn't tagged NCI - and why WUProp doesn't register actual runtime as supplied by the operating system !
I've applied your fix and look forward to more realistic stats :)

Judging by the errors in the logs, the fix is not accepted by Boinc
- not there
- not that
- boinc is not restarted

C:\ProgramData\BOINC\projects\boinc.gridcomputingcenter.org\app_config.xml
<app_config>
<app_version>
<app_name>monkeys_v1</app_name>
<avg_ncpus>0.001</avg_ncpus>
</app_version>
<app_version>
<app_name>monkeys_v2</app_name>
<avg_ncpus>0.001</avg_ncpus>
</app_version>
<app_version>
<app_name>monkeys_v3</app_name>
<avg_ncpus>0.001</avg_ncpus>
</app_version>
<app_version>
<app_name>monkeys_v4</app_name>
<avg_ncpus>0.001</avg_ncpus>
</app_version>
</app_config>

There is one more solution - to launch monkeys (NCI only) in a separate instance of Boinc with its copy of WUprop. Described in the news on the project server

Dirk Broer
     
Send message
Joined: 3 Nov 10
Posts: 38
Credit: 1,056,730
RAC: 692
Total hours: 2,227,389
Message 5108 - Posted: 18 May 2017, 16:22:26 UTC - in response to Message 5099.
Last modified: 18 May 2017, 16:23:55 UTC

When CERN consolidated their apps, for the most part things were swapped here without too much trouble.


I beg to differ here:
vLHCathome-dev
Sixtrack Simulation
ATLAS Simulation

lhcathome-dev new name for the former vLHCathome-dev
Sixtrack Simulation
ATLAS Simulation
CMS Simulation
Theory Simulation
LHCb Simulation
ALICE Simulation
Benchmark Application

ATLAS@home
ATLAS Simulation
ATLAS Simulation Running on Multiple Core

VirtualLHC@home
LHCb Simulations
CMS Simulation
Theory Simulations 64 bit
Theory Simulations

LHC@home new name for LHC@Home 1.0, incorporating ATLAS@home and VirtualLHC@home:
ATLAS Simulation
LHCb Simulation
CMS Simulation
Theory Simulation
sixtracktest
SixTrack
____________

HAL9000
 
Avatar
Send message
Joined: 14 Dec 11
Posts: 19
Credit: 5,688,288
RAC: 404
Total hours: 7,899,878
Message 5109 - Posted: 18 May 2017, 22:18:25 UTC - in response to Message 5092.
Last modified: 18 May 2017, 22:21:01 UTC

Check error log - "facteur correction: 0.500000 (4/8.010000)"

You ran 8 threads on 4 cores
Probably you have problems with monkeys, which under Windows are considered to be full CPU applications
Use the app_config to decrease the usage to 0.001 CPU

Due to the NCI/CPU app testing at the Monkeys project I had configured my app_config several months ago. However I was using a value of <avg_ncpus>0.000001</avg_ncpus> for the apps and after reading this thread I checks my stats. It seems with a value like that BOINC converts it to 1e-006. Which wuprop doesn't seem to understand. So it continued to count the 4 monkey apps as full CPU apps on my 4 CPU system and triggered "facteur correction".
The strange bit is that I only run 2 CPU, 1 GPU, and the 4 monkey apps on that system but still had "facteur correction: 0.500000 (4/8.010000)". As if 8 CPU apps were actually running. On a single CPU system running the 4 monkey apps I did see "facteur correction: 0.250000 (1/4.010000)". Which would be expected.

Changing the value to 0.001 looks to have corrected everything for now.
At least until they figure out how to set the plan class to nci for the nci applications.

NOTE: Restarting BOINC does seem to be required after setting <avg_ncpus>

Sergey Kovalchuk
     
Send message
Joined: 27 Oct 15
Posts: 111
Credit: 381,180
RAC: 260
Total hours: 1,261,137
Message 5111 - Posted: 19 May 2017, 5:43:56 UTC - in response to Message 5108.

When CERN consolidated their apps, for the most part things were swapped here without too much trouble.


I beg to differ here:
vLHCathome-dev
Sixtrack Simulation
ATLAS Simulation


LHC@home 1.0
sixtracktest

Beauty@LHC
Beauty@LHC virtual machine for physics simulations

vLHCathome-dev
Daily tricks with the project, when 5 applications for an hour fly into a deep minus

Dr Who Fan
   
Avatar
Send message
Joined: 29 Jul 11
Posts: 129
Credit: 386,545
RAC: 192
Total hours: 418,555
Message 5136 - Posted: 24 May 2017, 4:08:30 UTC

Looking over a spreadsheet I created to track my hours, etc here at WuProp...I also am NOT receiving A FULL HOUR FOR HOUR CREDIT FOR ANY OF MY PROJECTS FOR A WHILE HERE!

Appears the reduction in hours started possibly around 04/15 or 4/16/2017.

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:

EXAMPLE 1 > This PC is a 4 CORE but WuProp seems to think it might be an 8 core???
https://wuprop.boinc-af.org/result.php?resultid=71350668

Stderr output
<core_client_version>7.7.2</core_client_version>
<stderr_txt>
Erreur assignation taille wu
Erreur assignation taille application
13:46:39 (3232): facteur correction: 0.500000 (4/8.010000)

------

EXAMPLE 2 > This PC is a 2 CORE but WuProp seems to think it might be 6 cores???
https://wuprop.boinc-af.org/result.php?resultid=71353683
<core_client_version>7.7.2</core_client_version>
<stderr_txt>
Erreur assignation taille wu
Erreur assignation taille application
16:17:26 (2892): facteur correction: 0.333333 (2/6.010000)

-----

EXAMPLE 3 > This PC is a 2 CORE but WuProp seems to think it might be 6 cores???
https://wuprop.boinc-af.org/result.php?resultid=71353486
<core_client_version>7.7.2</core_client_version>
<stderr_txt>
Erreur assignation taille wu
Erreur assignation taille application
16:06:56 (2708): facteur correction: 0.333333 (2/6.010000)

------

EXAMPLE 4 > This PC is a 2 CORE but WuProp seems to think it might be 6 cores???
https://wuprop.boinc-af.org/result.php?resultid=71353156
<core_client_version>7.7.2</core_client_version>
<stderr_txt>
Device GeForce GT 630 already detected as NVIDIA GeForce GT 630
Erreur assignation taille wu
Erreur assignation taille application
15:51:23 (8028): facteur correction: 0.333333 (2/6.010000)

------

EXAMPLE 5 > This PC is a 2 CORE but WuProp seems to think it might be 6 cores???
https://wuprop.boinc-af.org/result.php?resultid=71348754
<core_client_version>7.7.2</core_client_version>
<stderr_txt>
Erreur assignation taille wu
Erreur assignation taille application
12:01:23 (1984): facteur correction: 0.333333 (2/6.010000)

____________

Sergey Kovalchuk
     
Send message
Joined: 27 Oct 15
Posts: 111
Credit: 381,180
RAC: 260
Total hours: 1,261,137
Message 5137 - Posted: 24 May 2017, 5:49:26 UTC - in response to Message 5136.

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:


Problem with BOINC client & Monkeys NCI

(2 or 4 core) / ((2 or 4 real CPU tasks) + 4 NCI monkeys + 0.01 WUprop) = 0.33 or 0.50

Use separate BOINC instance for monkeys NCI & main instance for monkeys CPU + other tasks

Dr Who Fan
   
Avatar
Send message
Joined: 29 Jul 11
Posts: 129
Credit: 386,545
RAC: 192
Total hours: 418,555
Message 5138 - Posted: 24 May 2017, 5:56:45 UTC - in response to Message 5137.
Last modified: 24 May 2017, 5:58:09 UTC

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:


Problem with BOINC client & Monkeys NCI

(2 or 4 core) / ((2 or 4 real CPU tasks) + 4 NCI monkeys + 0.01 WUprop) = 0.33 or 0.50

Use separate BOINC instance for monkeys NCI & main instance for monkeys CPU + other tasks


How do you Use separate BOINC instance for monkeys NCI?????

--- edit --- and how will that work for the BoincTasks monitering program??
____________

Sergey Kovalchuk
     
Send message
Joined: 27 Oct 15
Posts: 111
Credit: 381,180
RAC: 260
Total hours: 1,261,137
Message 5139 - Posted: 24 May 2017, 8:21:45 UTC - in response to Message 5138.
Last modified: 24 May 2017, 8:22:25 UTC

How do you Use separate BOINC instance for monkeys NCI?????

--- edit --- and how will that work for the BoincTasks monitering program??


GCC forum - http://boinc.gridcomputingcenter.org/forum_thread.php?id=205&postid=1758

Add --allow_remote_gui_rpc in command line - for "remote" control

Or same options to cc_config:
<allow_multiple_clients>1</allow_multiple_clients>
<allow_remote_gui_rpc>1</allow_remote_gui_rpc>


Or add 127.0.0.1 to remote_hosts.cfg

Use 127.0.0.1:9999 for boinc manager
Add new comp in BoincTask and set 127.0.0.1 to ip & 9999 to port


On GCC pref set only NCI apps for 2nd instance & only CPU apps for "old" host (1st instance)

mmonnin
   
Send message
Joined: 22 Aug 16
Posts: 96
Credit: 347,168
RAC: 568
Total hours: 1,105,211
Message 5141 - Posted: 24 May 2017, 13:41:03 UTC - in response to Message 5139.

How do you Use separate BOINC instance for monkeys NCI?????

--- edit --- and how will that work for the BoincTasks monitering program??


GCC forum - http://boinc.gridcomputingcenter.org/forum_thread.php?id=205&postid=1758

Add --allow_remote_gui_rpc in command line - for "remote" control

Or same options to cc_config:
<allow_multiple_clients>1</allow_multiple_clients>
<allow_remote_gui_rpc>1</allow_remote_gui_rpc>


Or add 127.0.0.1 to remote_hosts.cfg

Use 127.0.0.1:9999 for boinc manager
Add new comp in BoincTask and set 127.0.0.1 to ip & 9999 to port


On GCC pref set only NCI apps for 2nd instance & only CPU apps for "old" host (1st instance)


WUProp won't work on the 2nd instance w/o the app_config for it. I wrote a complete guide for Windows, Linux, WUProp, and BoincTasks.
http://www.overclock.net/t/1628924/guide-setting-up-multiple-boinc-instances/0_30

HAL9000
 
Avatar
Send message
Joined: 14 Dec 11
Posts: 19
Credit: 5,688,288
RAC: 404
Total hours: 7,899,878
Message 5145 - Posted: 25 May 2017, 0:48:20 UTC - in response to Message 5137.

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:


Problem with BOINC client & Monkeys NCI

(2 or 4 core) / ((2 or 4 real CPU tasks) + 4 NCI monkeys + 0.01 WUprop) = 0.33 or 0.50

Use separate BOINC instance for monkeys NCI & main instance for monkeys CPU + other tasks

It seems there are at least 3 workarounds to have the hours counted correctly until the Monkeys apps are set correctly.
1) Make an app_config for the monkeys project
2) Set the number of CPUs in BOINC higher
3) Run a separate instances of BOINC

Profile STE\/E
         
Avatar
Send message
Joined: 28 Mar 10
Posts: 360
Credit: 3,059,054
RAC: 1,127
Total hours: 16,117,672
Message 5158 - Posted: 28 May 2017, 10:54:56 UTC - in response to Message 5145.

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:


Problem with BOINC client & Monkeys NCI

(2 or 4 core) / ((2 or 4 real CPU tasks) + 4 NCI monkeys + 0.01 WUprop) = 0.33 or 0.50

Use separate BOINC instance for monkeys NCI & main instance for monkeys CPU + other tasks

It seems there are at least 3 workarounds to have the hours counted correctly until the Monkeys apps are set correctly.
1) Make an app_config for the monkeys project
2) Set the number of CPUs in BOINC higher
3) Run a separate instances of BOINC


I don't think it really matters what you do, WuProp is only going to give you credit for Hr's run according to CPU Usage. In other words if you run 1 Project on a 4 Core CPU in HT Mode (8 Cores) for the 6 Hr's each WuProp Wu runs you'll only get about 32 Hr's of Credit at WuProp for the 6 Hr's ran.

This gets reduced even further if you run 2 BOINC Clients, you'll only get about 60 Hr's credit (30 for each Client)for the 2 Clients for each 6 Hr's of run time.

mmonnin
   
Send message
Joined: 22 Aug 16
Posts: 96
Credit: 347,168
RAC: 568
Total hours: 1,105,211
Message 5159 - Posted: 29 May 2017, 1:57:40 UTC - in response to Message 5158.

Seems to be a PROBLEM WITH HOW WuProp IS CALULATING ONLY ~33 to 50% of REPORTED CPU RUN-TIMES as shown below in some recent completed WuProp Task:


Problem with BOINC client & Monkeys NCI

(2 or 4 core) / ((2 or 4 real CPU tasks) + 4 NCI monkeys + 0.01 WUprop) = 0.33 or 0.50

Use separate BOINC instance for monkeys NCI & main instance for monkeys CPU + other tasks

It seems there are at least 3 workarounds to have the hours counted correctly until the Monkeys apps are set correctly.
1) Make an app_config for the monkeys project
2) Set the number of CPUs in BOINC higher
3) Run a separate instances of BOINC


I don't think it really matters what you do, WuProp is only going to give you credit for Hr's run according to CPU Usage. In other words if you run 1 Project on a 4 Core CPU in HT Mode (8 Cores) for the 6 Hr's each WuProp Wu runs you'll only get about 32 Hr's of Credit at WuProp for the 6 Hr's ran.

This gets reduced even further if you run 2 BOINC Clients, you'll only get about 60 Hr's credit (30 for each Client)for the 2 Clients for each 6 Hr's of run time.


Then how do multiple GPU tasks at once work? I get 48 hours of Moo per day on a single GPU since I run 2 tasks at once.

Profile STE\/E
         
Avatar
Send message
Joined: 28 Mar 10
Posts: 360
Credit: 3,059,054
RAC: 1,127
Total hours: 16,117,672
Message 5160 - Posted: 29 May 2017, 3:16:04 UTC

Then how do multiple GPU tasks at once work? I get 48 hours of Moo per day on a single GPU since I run 2 tasks at once


WuProp may think each task is running at 100% so it gives full Hr's for the work done.

mmonnin
   
Send message
Joined: 22 Aug 16
Posts: 96
Credit: 347,168
RAC: 568
Total hours: 1,105,211
Message 5162 - Posted: 29 May 2017, 12:06:10 UTC - in response to Message 5160.

Then how do multiple GPU tasks at once work? I get 48 hours of Moo per day on a single GPU since I run 2 tasks at once


WuProp may think each task is running at 100% so it gives full Hr's for the work done.


The same is applied to CPUs. A task is a task is a task according to WUProp.

Profile STE\/E
         
Avatar
Send message
Joined: 28 Mar 10
Posts: 360
Credit: 3,059,054
RAC: 1,127
Total hours: 16,117,672
Message 5163 - Posted: 29 May 2017, 15:02:52 UTC - in response to Message 5162.

Then how do multiple GPU tasks at once work? I get 48 hours of Moo per day on a single GPU since I run 2 tasks at once


WuProp may think each task is running at 100% so it gives full Hr's for the work done.


The same is applied to CPUs. A task is a task is a task according to WUProp.


It should be but it isn't, I run 2 instances of BOINC & run my 4 Core CPU's in HT Mode so in actuality I should be getting credit for running 16 Wu's @ once. But according to BOINCtasks the CPU Usage for each of the 16 Thread's is all over the Board, may be as little as 2% or as high as 99%, so I figure WuProp gives you Hr's according to the CPU Usage.

It's the only reason I can see getting only 60 Hr's instead of the 96 Hr's (16x6) for 2 WuProp Wu's turned in ...

Dirk Broer
     
Send message
Joined: 3 Nov 10
Posts: 38
Credit: 1,056,730
RAC: 692
Total hours: 2,227,389
Message 5331 - Posted: 6 Jul 2017, 7:21:50 UTC
Last modified: 6 Jul 2017, 7:23:48 UTC

It is getting worse with NCI applications, they're being treated like they're full-blown CPU applications.

When you have a single-core machine (e.g. a Raspberry Pi B+) and ran all six QCN (NCI) projects you used to get 24 hours per project per day. Of course you also run WUProp (NCI) and when Goofyx started, four Goofyx (NCI) applications as well. That used to be 6+1+4=11 (NCI) apps that ran 24 hrs/day and each was counted as such in WUProp.

But not anymore so. If you do that now, each of those apps will only give you 24/11 hrs/day. That less than 3 hrs/day, despite they're all running 24/7!
____________

1 · 2 · 3 · Next
Post to thread

Message boards : Number crunching : Only half (50%) elapsed time registered


Home | My Account | Message Boards | Results


Copyright © 2017 Sebastien