I ask because it's not happening on my machines, unless I misunderstand the concept of switching between applications.

I run three projects - WCG, Virtual Prairie and Rosetta, and I set my machines to "switch between applications", expecting that each project gets a fair slice of time. Thus far it hasn't happened, no matter how I tweak settings in clients or BAM.

Am I missing something here?

(I should add that I'm using version 6.10.58 on Windoze Vista)

Do they have equal resource share? Although I'm not an expert on this topic, I've found that predicting what applications will run when can become very complicated. In addition to resource share, "credit debt" will pile up when a project hasn't done any work for a while. That, and moon phase, ambient temperature, mood of the operators, etc.

And checkpointing! An application that hasn't written a checkpoint won't get switched out if you have "keep in memory" set to "no".

@wertperch: What's your time setting for "Switch between tasks every..."?

Gruß,
Gundolf

Equal resource share, no. Although operator mood has been pretty darned black recently. Cab software get scared so badly it wants to hide in its kennel?

Checkpointing? So what should that "keep in memory" setting be?

And the time setting is currently 45 minutes, but it's been everything from 5 minutes to 5 hours.

Most BOINC applications write information they need to restart after an interrupt into an external file at certain points in time/calculation, called checkpoints. If those checkpoint information is not written, BOINC knows not to remove that application from memory, lest it has to start over.


It's in your computing preferences:

Leave tasks in memory while suspended?           no

Suspended tasks will consume swap space if 'yes'

And you really don't see some tasks getting swapped out from time to time (Status: Waiting to run)?

Gruß,
Gundolf

Yes, but this has actually only started happening in the past few hours, and now I am even more confused.

The screenshot I've posted at http://picasaweb.google.com/wertperch/DropBox#5519776295791071602 shows that "Help Conquer Cancer" is split between running and "waiting to run", the Rosetta has been stuck at that percentage for almost 48 hours, and seems destined never to run again.

The software certainly is not behaving as I would expect - why not finish one round of "HCC" before starting another? When the heck is the Rosetta going to get another chance? That's been stuck at 1 hour elapsed for almost three days.

I've made your link clickable.

BOINC 6.10.58 seems sometimes to have problems with starting tasks before deadline. To check that you'd have to enable some logging flags. Look at the BOINC Message boards for information and help.

Or you could try if 6.10.56 (or earlier) works better.

Gruß,
Gundolf

Has Gundolf's suggestion of using 6.10.56 (or earlier) resolved your issue?
Looking at the Report deadline's of your screen shot, it appears as if a lot of crunching has to be done before Rosetta kicks in again.
I am not sure wether the deadlines are taken into consideration by the manager when switching between tasks, irespective of the settings.
On some of my older PC's (10+years old and slow crunchers) I tend to use the Report deadline as a guide and manipulate the tasks by manualy suspending the ones that are running and not switching according to the settings in a way that I thought they should.
I am sure that this is not the idea of the BOINCmanager but it mostly works to compleet tasks in time.

Well, thank you all for your suggestions - now things are switching, but I remain confused as to how the software prioritises tasks - I have a work unit that has been stuck at 98% for a couple of days, and I've noticed that each time I suspend a project to try and clear things up, it seems to get more confusing.

Maybe patience is the thing I need, rather than kicking the settings...?

Try setting all projects to no new work or no new tasks before you suspend any (excluding of course the project on which you wish to force a completion). This should prevent work units from downloading from all projects and starting before the completion of your desired work unit. You should also, perhaps, set your switch time to one hour and leave it there for a few days. BOINC is most likely attempting to keep up with a frequent change of input and this may be the cause of some of your problems. In general, I believe a project will tend to run to it's next checkpoint and then switch to another project. If this checkpoint takes 3hrs. to reach, you would then "owe" another project(s) 2hrs of crunch time to even things out.

Your short and long debts must be way off and by fiddling around by manual intervening it could be become even worse.
First think about the resource share you gave to the projects. Are they what you want to they should be?

Do not load too much work.
If you are always connected to the internet set in Boinc preferences the network setting "connect every days" to 0 (zero) and the additional work buffer to 1 day at the most. If you are connecting to a project for the first time on a machine, it's wise to lower this to e.g. 0.01 to see what the first task will do and to set the first DCF (duration correction factor), before asking too much work.

To correct the LTD and STD (long and short term debts), put in the options part of your cc_config.xml " <zero_debts>1</zero_debts>"
and restart your boinc client. After the restart you could change it to <zero_debts>0</zero_debts> to avoid resetting after every restart.

Good luck!

This is starting to make sense now, I realise that I've messed things up by a) keeping too large a work buffer, and 2) fiddling with things. I'll try to clear these debts and take your sound advice!

Pardon my ignorance, but if I do this, will I lose work that's already in progress? I'm still unsure what this setting does.

It depends on what projects you are running. If you have running projects without checkpoints, they would start from the beginning.
First thing you could do is to put <checkpoint_debug>1</checkpoint_debug> in the log_flags part of the cc_config.xml file and
after that press "read config file" in the Advanced menu of your Boinc Manager.

In the messages you can see now whether and how often checkpoints are written. Up to you to decide how much work should done from scratch or from the last checkpoint.

Did that - what I got out was this:

9/26/2010 14:05:12 rosetta@home [cpu_sched_debug] Project has 2 projected CPU deadline misses
9/26/2010 14:05:12 boincsimap [cpu_sched_debug] Project has 22 projected CPU deadline misses

So, now what?

I'm missing the checkpoint_debug flag in your message.

Should look like this:

26/09/2010 13:36:36 log flags: file_xfer, sched_ops, task, checkpoint_debug

Is it in the flags part of the file?

E.g.
<cc_config>
<log_flags>
<checkpoint_debug>1</checkpoint_debug>
</log_flags>
<options>
<zero_debts>0</zero_debts>
</options>
</cc_config>