In order to satisfy the requirements of the 3 different projects (CPDN, S@H & Predictor), I've spent a nice loooooong time looking at the storage and CPU (time to complete WU's) requirements and compared that against the specifications of the 7 computers that I have running BOINC.

2 of them are at home, they're similar to each other, so I set them as "Home". (duh, eh? ;))

3 of them are resonably fast, have fairly good sized HD's and at work, so I set them to "Work".

2 of them are slower CPU's, with smaller HD's & are also at work, but I set them as "School" so I could allocate a larger % of time to CPDN and a larger % of HD space that CPDN needs.

Make sense so far? :)

I set the settings like this:

Home: Disk 16% (Max 10Gb) CPDN 8%, S@H 90%, Pred 2%
Work: Disk 16% (Max 4Gb) CPDN 22%, S@H 76%, Pred 2%
School: Disk 24% (Max 2Gb) CPDN 62%, S@H 36%, Pred 2%

(as you can see, I really only have predictor as a "fallback" in case neither of the other two can get work)

I tried to update the projects on the various computers, but as soon as I did, they would all revert to "Home" in the Preferences of each site that I would choose. I tried setting the 'Default' site differently, but it made no difference. All I really know with 100% certainty is: 'Push Update on client = setting location as Home on project site'

So I gave up trying to do that (after resetting the locations of the machines dozens of times trying to figure the problem out).

Instead, I waited patiently for each client to contact the schedulers on their own and checked back daily to see changes.

The disk allocations (in total) appear to have worked perfectly. Each client machine is allocating a total HD space just as I'd like it.

However, I'm still faced with a situation that nearly every client is reporting an allocation to each project that doesn't even come CLOSE to what I specified.

I'm not home, so I'm not positive of those #'s, but the "Work" machines are 42.97% CPDN, 56.6% S@H and ~.5% Predictor and the "School" machines are 40.52% CPDN, 58.82% S@H and ~.5% Predictor!

I've tried adding up the default resource share and including that, I've tried every mathematical computation I can think of to explain where the clients are getting these %'s from and they don't appear to relate to anything I set!

I'm tired, frustrated and completely out of ideas after trying to investigate and figure this out from hours upon hours of work and experimentation. Can anyone help?

CPU Scheduling in Ver 4.05
Client scheduling policies


M7 Seti@h Berkeley's Staff Friends Club ©

> CPU Scheduling in Ver
> 4.05
> Client scheduling
> policies
>
>
> M7 Seti@h Berkeley's Staff Friends Club ©
>

I suppose that's better than nothing, but I've already read both of those documents. The first doesn't have a thing to do with what I'm talking about and the second states:

"Enforce resource shares. The ratio of CPU time allocated to projects that have work, in a typical period of a day or two, should be approximately the same as the ratio of the user-specified resource shares. If a process has no work for some period, it does not accumulate a 'debt' of work."

Should be approximately the same as the ratio of the user-specified resource shares?? Hrmm, okay... so what if they're not?

When I have two clients... one should be approximately 90/10 ratio and the other should be approximately 30/70 in the other direction and yet they're both giving the same, doesn't that mean something is wrong?

Let me boil this down: General resource sharing do everything. Resource sharing fo' skool, home and werk dun do nuthin'? I too dum to reed tecknikal stuffs. Sumbody help?

There are bugs in the venue specific preferences. Until they get sorted out you may notice some odd things happening if you use them.

The settings for the most part translate across computers with different capabilities fairly well. Unfortunately the settings you desire to be different are not ones that translate well, with one exception; if the computer does not meet the project specified minimums it will not get work from that project.

John Keck -- BOINCing since 2002/12/08 --

> There are bugs in the venue specific preferences. Until they get sorted out
> you may notice some odd things happening if you use them.

Okay, that makes sense. Now that I'm finding some other posts, I can see that "bugs" is an understatement. Looks like the location settings are rather like the 'Door Close' button on elevators... completely non-functional, but put there to give harried yuppies (or in this case, IT personnel) something to waste their time on. :P


> The settings for the most part translate across computers with different
> capabilities fairly well. Unfortunately the settings you desire to be
> different are not ones that translate well, with one exception; if the
> computer does not meet the project specified minimums it will not get work
> from that project.

I've had no trouble getting work from projects (once I figured out that the general settings were the only effective ones and the others were non-functional). BTW, I'm getting a bit of a chuckle from the way everyone soft-pedals the problems in BOINC. In this case "are not ones that translate well" = "ones that do nothing".

Sigh... I can't think of anything to do other than leave them set the way I want them to be and then I suppose I'll know the feature will be functional when I see them start behaving correctly someday. :\

Whee!! It just keeps getting better and better! My home machines are both still following the General settings, but I came in to work to find that one of my machines was running 90% SAH! My first (for about .4 seconds) reaction, was GREAT! it finally paid attention to the setting by location! Then I scratched my head and thought.... but that's not the settings for a "Work" machine.

I brought up my account and sure enough, the machine had changed it's location to "Home" again, so I changed it back.... again.

Also, one of the "School" machines changed it's location back to "Home", but continued to follow the general resource share settings.

Is anyone else having this level of 'randomness' to how the resource sharing works?

Sincerely, yours in confusion,

Thunder

That same client had to connect to get more work....

did the same thing again. Switched it's own location back to "Home".

Anyone have a solution to stop this?

> That same client had to connect to get more work....
>
> did the same thing again. Switched it's own location back to "Home".
>
> Anyone have a solution to stop this?
>
Wait for the developers to fix the web pages, and possibly the CC.

4.05 appears to count WUs rather than projects when calculating resource shares. For example if you had 50 CPDN and 50 S@H, and you have 3 S@H WUs and 1 CPDN, the appearance is that you would bet 25 CPDN and 75 S@H for shares. This appears to be fixed in 4.08.

> 4.05 appears to count WUs rather than projects when calculating resource
> shares.

Actually, the couple of machines that I've somehow gotten to correctly set their location are resource sharing perfectly based on MY preferences and despite having only 1 CPDN and ~6-20 S@H WU's, it does appear to be spending roughly the amount of time-slicing it should be on each based on those preferences. I can see that there are a lot of factors in the time-slicing algorithms, so I would anticipate that if I had WU's that were near expiring or something similar, that could change, of course.

Thanks for the info you provided. I can understand if the BOINC dev folks don't have the location feature working correctly, but I certainly wish they could post some sort of warning on the pages where you set it that it's still in development or something similar. :(

I'm going to try editing the client_state.xml as mentioned in this thread: http://setiweb.ssl.berkeley.edu/forum_thread.php?id=2796 and see if the setting 'sticks' or not.

I'll report back once I'm confident I know what it's doing. ;)