climateprediction.net home page
limiting number of simultaneous work units

limiting number of simultaneous work units

Questions and Answers : Unix/Linux : limiting number of simultaneous work units
Message board moderation

To post messages, you must log in.

AuthorMessage
Richard Kimber

Send message
Joined: 17 Oct 06
Posts: 5
Credit: 209,102
RAC: 0
Message 40682 - Posted: 16 Sep 2010, 12:26:25 UTC
Last modified: 16 Sep 2010, 12:26:53 UTC

How may I ensure that I only run one work unit at a time? Recently I have been getting two that, because of the long computation times, are both run at high priority, thus excluding work on other projects.
ID: 40682 · Report as offensive     Reply Quote
Profile tullio

Send message
Joined: 6 Aug 04
Posts: 264
Credit: 965,476
RAC: 0
Message 40684 - Posted: 16 Sep 2010, 17:25:26 UTC - in response to Message 40682.  

How may I ensure that I only run one work unit at a time? Recently I have been getting two that, because of the long computation times, are both run at high priority, thus excluding work on other projects.

This depend on your cache limit. I have one of only 0.25 days each project and I only get a WU at a time.
Tullio
ID: 40684 · Report as offensive     Reply Quote
Richard Kimber

Send message
Joined: 17 Oct 06
Posts: 5
Credit: 209,102
RAC: 0
Message 40686 - Posted: 16 Sep 2010, 17:40:40 UTC - in response to Message 40684.  

I'm not sure I can see anything called a 'cache limit', but in my general Boinc preferences I have 'Maintain enough work for 0.25 days'. Is this it?

If so, I'm getting two work units despite that. I get more than one WU for other projects, but it's not a problem because the computation time is small in relation to the deadline.

By aborting one of the climate units, the other ceases to run at high priority, and everything is normal.
ID: 40686 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 40687 - Posted: 16 Sep 2010, 18:59:07 UTC
Last modified: 16 Sep 2010, 19:00:43 UTC

Hi Richard, welcome to the forum

Yes, the cache limit is the additional work buffer in one's Boinc manager Advanced Preferences.

How many new climate models you get depends to some extent on the resource share you've allocated to CPDN. You can check what it says in your Boinc manager Projects tab. Part of the Boinc program consists of a work scheduler. As its name suggests the scheduler decides how much work to fetch from a project, how to make sure the tasks fetched will complete before their deadlines, what order to run tasks from the different projects in and how to respect the resource share you've set for each project.

If you just let Boinc do its own thing and take these decisions for itself it will do all these things. It's designed to be left alone with no user intervention.

However, if the scheduler finds itself with a big load of climate models that need to be completed by a particular date, it may decide that the best way to fulfil all those goals is to run the climate models first. Let it do that. Your Boinc scheduler will then work out for itself that over the long term it's allocated too much time to CPDN and not enough to your other projects. CPDN has accumulated a debt. The scheduler will then, as soon as it can, give extra time to the other projects to even up their relative debts. In the long term you will find that Boinc will give the exact resource share to each project that you wanted. This could mean Boinc refusing to download new CPDN work for a while.

It honestly doesn't matter if Boinc gives more time to CPDN for a while and later gives lots of time to, say, Seti.

If this Boinc behaviour gets on your nerves you can if you wish in the Boinc manager Tasks tab select a model by clicking on it then press the Suspend button. It will stay suspended until you run it again. The Boinc scheduler will later take your intervention into account.

It's more helpful to CPDN if you suspend models rather than aborting them. If by chance suspending a model causes it to go over its deadline you don't need to worry. The CPDN servers (but not the servers from other projects) accept results you upload after the deadline; the researchers will still use overdue models. The Boinc scheduler doesn't know this, however. It thinks the climate models absolutely must complete by their deadline, which explains why sometimes it puts them into high priority mode.

Hope that helps.
Cpdn news
ID: 40687 · Report as offensive     Reply Quote
Richard Kimber

Send message
Joined: 17 Oct 06
Posts: 5
Credit: 209,102
RAC: 0
Message 40692 - Posted: 17 Sep 2010, 14:22:13 UTC - in response to Message 40687.  


If you just let Boinc do its own thing and take these decisions for itself it will do all these things. It's designed to be left alone with no user intervention.

However, if the scheduler finds itself with a big load of climate models that need to be completed by a particular date, it may decide that the best way to fulfil all those goals is to run the climate models first. Let it do that. Your Boinc scheduler will then work out for itself that over the long term it's allocated too much time to CPDN and not enough to your other projects. CPDN has accumulated a debt. The scheduler will then, as soon as it can, give extra time to the other projects to even up their relative debts. In the long term you will find that Boinc will give the exact resource share to each project that you wanted. This could mean Boinc refusing to download new CPDN work for a while.

It honestly doesn't matter if Boinc gives more time to CPDN for a while and later gives lots of time to, say, Seti.

Thanks for your reply, and I now see what Boinc is trying to do. However, I prefer a more even style of work. As things stand, it seems I could spend a couple of months on high priority climate models at the expense of projects that take maybe an hour, and find that when the climate models are done there's no work available on the other projects. I guess your suggestion of suspending one of the climate models is the answer. Though I think it would be much better if the user could control the number of models per project.
ID: 40692 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 40695 - Posted: 17 Sep 2010, 14:53:55 UTC
Last modified: 17 Sep 2010, 15:02:54 UTC

Boinc crunchers have asked for the option of just getting one task at time but the Boinc programmers seem to have decided that this wouldn't sit well with the other things the Boinc scheduler does.

BTW, a new batch of slab models called HadSM has just been launched. They're longer than FAMOUS and HadAM3P regional so you'd do well to go into your account, then the climateprediction preferences, then edit the model types you want to exclude this. Say you want HadAM3P regional and FAMOUS only. And of the three regionals, the SA (Southern Africa) is the shortest.

You can also micromanage whether you get work from your various projects by using the No new tasks button in the Boinc manager Projects tab.

As I said, the Boinc scheduler is designed for set-and-forget crunchers but there's nothing to stop people using all the options to get what they want. It's designed for interventionists as well.
Cpdn news
ID: 40695 · Report as offensive     Reply Quote
Richard Kimber

Send message
Joined: 17 Oct 06
Posts: 5
Credit: 209,102
RAC: 0
Message 40703 - Posted: 17 Sep 2010, 22:17:42 UTC - in response to Message 40695.  

Thanks for your helpful advice.
ID: 40703 · Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : limiting number of simultaneous work units

©2021 climateprediction.net