Thread 'no GPU work fetched whn fetch minimal flag set'
Message boards : GPUs : no GPU work fetched whn fetch minimal flag set
Message board moderation
| Author | Message |
|---|---|
 jay_eSend message Joined: 8 Mar 07 Posts: 115 
|
Hi, I see this on 2 systems windows vista with BOINC 7.4.42 Ubuntu Linux with BOINC 7.6.2 In both event logs, I see: 7/2/2015 11:04:56 PM | SETI@home | Sending scheduler request: Requested by user. 7/2/2015 11:04:56 PM | SETI@home | Not requesting tasks 7/2/2015 11:04:58 PM | SETI@home | Scheduler request completed except the other machine is running on Einstein GPU WU - but the 'not requesting' is the same. I had changed the cc_config.xml in both to:
<cc_config>
<log_flags>
<file_xfer>1</file_xfer>
<sched_ops>1</sched_ops>
<task_debug>1</task_debug>
</log_flags>
<options>
<fetch_minimal_work>1</fetch_minimal_work>
<max_file_xfers>3</max_file_xfers>
<max_file_xfers_per_project>1</max_file_xfers_per_project>
<report_results_immediately>1</report_results_immediately>
<save_stats_days>60</save_stats_days>
<start_delay>60</start_delay>
</options>
</cc_config>
I was getting aggravated with one project getting only 1 wu and the other project getting several. Thought this config would resolve. Anyway. The side effect was that both machines did not get any GPU work. The app_configs were set to 1.0 cpu and 1.0 gpu. So I let the CPU WUs complete - thinking that was the problem. Nope. With just 2 WU on a 4-kernel machine, running 100% of kernels, no GPU work fetched. I turned on the work_fetch_debug log flag... 7/2/2015 11:53:57 PM | | [work_fetch] ------- start work fetch state ------- 7/2/2015 11:53:57 PM | | [work_fetch] target work buffer: 180.00 + 864.00 sec 7/2/2015 11:53:57 PM | | [work_fetch] --- project states --- 7/2/2015 11:53:57 PM | DENIS@home | [work_fetch] REC 37.593 prio -0.000 can't request work: "no new tasks" requested via Manager 7/2/2015 11:53:57 PM | FiND@Home | [work_fetch] REC 63.548 prio -0.000 can't request work: "no new tasks" requested via Manager 7/2/2015 11:53:57 PM | World Community Grid | [work_fetch] REC 458.282 prio -0.076 can't request work: "no new tasks" requested via Manager 7/2/2015 11:53:57 PM | SETI@home | [work_fetch] REC 23490.307 prio -1.000 can request work 7/2/2015 11:53:57 PM | | [work_fetch] --- state for CPU --- 7/2/2015 11:53:57 PM | | [work_fetch] shortfall 2055.60 nidle 1.00 saturated 0.00 busy 0.00 7/2/2015 11:53:57 PM | DENIS@home | [work_fetch] share 0.000 7/2/2015 11:53:57 PM | FiND@Home | [work_fetch] share 0.000 7/2/2015 11:53:57 PM | World Community Grid | [work_fetch] share 0.000 [b]7/2/2015 11:53:57 PM | SETI@home | [work_fetch] share 0.000 blocked by project preferences[/B] 7/2/2015 11:53:57 PM | | [work_fetch] --- state for AMD/ATI GPU --- 7/2/2015 11:53:57 PM | | [work_fetch] shortfall 1044.00 nidle 1.00 saturated 0.00 busy 0.00 7/2/2015 11:53:57 PM | DENIS@home | [work_fetch] share 0.000 no applications 7/2/2015 11:53:57 PM | FiND@Home | [work_fetch] share 0.000 no applications 7/2/2015 11:53:57 PM | World Community Grid | [work_fetch] share 0.000 no applications 7/2/2015 11:53:57 PM | SETI@home | [work_fetch] share 1.000 7/2/2015 11:53:57 PM | | [work_fetch] ------- end work fetch state ------- 7/2/2015 11:53:57 PM | SETI@home | [work_fetch] set_request() for AMD/ATI GPU: ninst 1 nused_total 0.00 nidle_now 1.00 fetch share 1.00 req_inst 1.00 req_secs 1044.00 7/2/2015 11:53:57 PM | SETI@home | [work_fetch] request: CPU (0.00 sec, 0.00 inst) AMD/ATI GPU (1044.00 sec, 1.00 inst) 7/2/2015 11:53:57 PM | SETI@home | Sending scheduler request: To fetch work. I did not understand the entry with SETI "blocked by project preferences"... I set the fetch minimal back to zero; waited for backoff to wind down. Did an update. The Linux machine fetched Einstein GPU WU. The windows machine fetched SETI GPU WU. after a read of config files and the backoff completed... I forgot where to search the BOINC problem 'trac', but I did not see this in the gpu forums. Problem report needed? Thanks, Jay ID: 62890 ·  |
|
Send message Joined: 23 Apr 07 Posts: 1112 
|
7/2/2015 11:53:57 PM | | [work_fetch] ------- start work fetch state ------- That means the 'Use CPU' project preference at Setiathome is set to 'No'. Claggy ID: 62896 · |
Jord Send message Joined: 29 Aug 05 Posts: 15570 
|
It also helps to know what the meaning is of <fetch_minimal_work>1</fetch_minimal_work> It means: get one task per hardware device and then, until you restart BOINC, never again. It's one for testing purposes, where a project admin can test that his newly built scheduler and or plan_class gives work out to the intended hardware devices without having to store a full cache of work. One of these days I'll edit the description of it in the Wiki to say this, as I understand that what it now says may be confusing (I did not write that). ID: 62897 · |
|
jay_e Send message Joined: 8 Mar 07 Posts: 115
|
Thanks to Claggy and Jord!! Now I understand what fetch_minimal_work means! Jay ID: 62935 · |
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.