Configure Client for Minimum WU download (1 per CPU-Core)

Message boards : BOINC client : Configure Client for Minimum WU download (1 per CPU-Core)
Message board moderation

To post messages, you must log in.

AuthorMessage
Seth

Send message
Joined: 19 Nov 16
Posts: 63
Australia
Message 74883 - Posted: 21 Dec 2016, 13:38:11 UTC

Hi,
Whats the correct config in cc_config.xml to limit my client to the minimum amount of work to be downloaded. I only really want one WU per CPU downloaded at a time.

On my Mac client this is managed really well by setting "Storea at Leat" to Zero and "Store up to an additional" to zero.

For linux have tried setting <fetch_minimal_config>i</fetch_minimal_config> but this does not seem to work.

Need this for my grid.

Cheers
Seth
ID: 74883 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 74884 - Posted: 21 Dec 2016, 15:55:52 UTC - in response to Message 74883.  

For linux have tried setting <fetch_minimal_config>i</fetch_minimal_config> but this does not seem to work.

How did you set it? In cc_config.xml in the following usage?
<cc_config>
<log_flags>
</log_flags>
<options>
<fetch_minimal_config>1</fetch_minimal_config>
</options>
</cc_config>

This will fetch just one task per CPU and GPU. It will not fill the CPU with one task per core, it's just one task for the CPU, no matter how many cores.

Where did you place the cc_config.xml file?
Did you exit & restart BOINC afterwards?
ID: 74884 · Report as offensive
Seth

Send message
Joined: 19 Nov 16
Posts: 63
Australia
Message 74891 - Posted: 21 Dec 2016, 21:23:11 UTC - in response to Message 74884.  

Hi Jord,
From what I have read on this config item it fetchs one item, process it and then doesn't fetch anything else. Are you saying the behavior is different?

Also I want one task per core, not one per CPU.

Right now I'm experimenting with the file global_prefs_override.xml

In it I have set:
<work_buf_min_days>0.000000</work_buf_min_days>
<work_buf_additional_days>0.000000</work_buf_additional_days>

But still waiting for my linux client to work through the wu it had already downloaded first to see it in full effect. It has certainly stopped pulling new work while it process the backlog. When I have these values set on my mac client through the desktop app it works really well downloading new WU when an existing WU is reaching about 70% completion meaning there is always work ready just in time.

Cheers
Seth
ID: 74891 · Report as offensive
ChristianB
Volunteer developer
Volunteer tester

Send message
Joined: 4 Jul 12
Posts: 321
Germany
Message 74892 - Posted: 21 Dec 2016, 22:10:39 UTC

Hi Seth,

yes the <fetch_minimal_config> tag is not what you want as it only fetches work one time and you need to restart the client in order to get new work. This was implemented to allow BOINC to be used in HTCondor backfill jobs. But this has other problems that come with it.
If you just want to have a way to run BOINC in the background of your Cluster to avoid the spikes in power consumptions when the Cluster is not fully utilized it is better to use the work buf settings either in a global_prefs_override.xml or in the regular computing preferences of the project you are using. Preferences are synched if you have multiple projects so this is not a problem. You just need to make sure you use a stable project and maybe talk to the admins about your use-case and what it entails. Using this technique you will always have a certain number of tasks in progress and when you are using the Cluster 100% for yourself the BOINC jobs will starve and fail with EXIT_TIME_LIMIT_EXCEEDED in the end. This is not bad if the limit is set reasonable.

Btw: The Mac Client and the Linux Client should behave the same way if they do not than there is a problem somewhere.
ID: 74892 · Report as offensive
Seth

Send message
Joined: 19 Nov 16
Posts: 63
Australia
Message 74899 - Posted: 22 Dec 2016, 10:46:00 UTC - in response to Message 74892.  

Thanks Christian. But I have another problem now with my cluster. As I've configured the BOINC clients so they only have very few jobs, it means that update requests have become more frequent. Multiply this by the number of nodes I have running boinc I'm not getting a lot of resposnses from the server stating:

Not sending work - last request too recent

I'm guessing this is exacerbated by the fact all of my nodes are using the same BOIN project user ID?

Any suggestions on how I can disable this throttle on the server?

Cheers
Seth
ID: 74899 · Report as offensive
Seth

Send message
Joined: 19 Nov 16
Posts: 63
Australia
Message 74900 - Posted: 22 Dec 2016, 12:32:32 UTC - in response to Message 74899.  

Found it. I set <min_sendwork_interval> to zero.

Cheers
Seth
ID: 74900 · Report as offensive

Message boards : BOINC client : Configure Client for Minimum WU download (1 per CPU-Core)

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.