computer mood swings

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0
Topic 197393

One of my boxes seems to be bipolar. For example, it just came out of over a month of doing no Einstein at all and downloaded 108 new tasks over the course of a few hours early this morning. My other one maintains a nice balance of Einstein and Seti according to the projects' resource shares. What gives?

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2963965685
RAC: 711382

computer mood swings

I take it that we're talking about host 4124638?

Yes, that does seem to be an example of a rare but annoying fault in BOINC.

Particularly if we were able to sort the tasks into order of issue time, I think we'd see that a block of 2/3/4 tasks was allocated every minute for an hour or more - your event log should display the same pattern in more detail.

In the event log, you'll probably be able to see (depending on the log settings) that your BOINC Client is requesting the same modest amount of work, as quickly as it's allowed to, and failing to realise that new work is being allocated and it doesn't actually need any more.

This is, for certain, a (very rare) client problem, and nothing to do with the Einstein project. It just shows up here at Einstein more often than other projects, because Einstein is one of the few projects which can consistently supply that much work, as quickly as that.

I've reported this problem to the BOINC client developers before, but with no effect. I'll use you example to reinforce the point: it would be helpful if you could dig out and post the relevant part of your event log (or stdoutdae.txt) as backup evidence.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: I take it that we're

Quote:
I take it that we're talking about host 4124638?


Yes, that one. If it matters, it's 5947619 at Seti.

Quote:

Yes, that does seem to be an example of a rare but annoying fault in BOINC.

Particularly if we were able to sort the tasks into order of issue time, I think we'd see that a block of 2/3/4 tasks was allocated every minute for an hour or more - your event log should display the same pattern in more detail.

In the event log, you'll probably be able to see (depending on the log settings) that your BOINC Client is requesting the same modest amount of work, as quickly as it's allowed to, and failing to realise that new work is being allocated and it doesn't actually need any more.

This is, for certain, a (very rare) client problem, and nothing to do with the Einstein project. It just shows up here at Einstein more often than other projects, because Einstein is one of the few projects which can consistently supply that much work, as quickly as that.

I've reported this problem to the BOINC client developers before, but with no effect. I'll use you example to reinforce the point: it would be helpful if you could dig out and post the relevant part of your event log (or stdoutdae.txt) as backup evidence.


I'll see what I can do. It has continued to download more Einstein work and let its Seti tasks on board fall below the server limits.

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

archae86
archae86
Joined: 6 Dec 05
Posts: 3158
Credit: 7235254348
RAC: 1207128

N9JFE wrote:let its Seti

N9JFE wrote:
let its Seti tasks on board fall below the server limits.

What do you mean by that?

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2963965685
RAC: 711382

RE: N9JFE wrote:let its

Quote:
N9JFE wrote:
let its Seti tasks on board fall below the server limits.
What do you mean by that?


SETI has project-specific limits of "maximum 100 GPU tasks in progress" (and the same for CPU tasks, making a limit of 200 tasks in total).

This will simply be BOINC evening-out the historic resource share: if SETI has received the lion's share of the computer's time in the past, then BOINC will preferentially fetch from Einstein for a while to level the playing field.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: N9JFE wrote:let its

Quote:
N9JFE wrote:
let its Seti tasks on board fall below the server limits.
What do you mean by that?


Seti has a limit of 100 CPU tasks and 100 GPU tasks per host. Most of the time, even modestly decent crunchers like the one in question are constantly asking for more work and being refused because they're up to the limit. Right now, it's letting itself get below 100 Seti CPU tasks because it has so much Einstein.

(I only allow my other box to use its GPU for Boinc, so it sits at 100 Seti tasks, plus usually a few Einsteins. Sometimes it lets itself run dry of Einstein for a few days, but it never gorges itself.)

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: RE: N9JFE wrote:let

Quote:
Quote:
N9JFE wrote:
let its Seti tasks on board fall below the server limits.
What do you mean by that?

SETI has project-specific limits of "maximum 100 GPU tasks in progress" (and the same for CPU tasks, making a limit of 200 tasks in total).

This will simply be BOINC evening-out the historic resource share: if SETI has received the lion's share of the computer's time in the past, then BOINC will preferentially fetch from Einstein for a while to level the playing field.


If I thought that was all that was happening, I wouldn't worry about it. But it seems to take huge doses of Einstein at a time, run them preferentially because of their shorter deadlines, and then do none for a while.

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2963965685
RAC: 711382

RE: RE: RE: N9JFE

Quote:
Quote:
Quote:
N9JFE wrote:
let its Seti tasks on board fall below the server limits.
What do you mean by that?

SETI has project-specific limits of "maximum 100 GPU tasks in progress" (and the same for CPU tasks, making a limit of 200 tasks in total).

This will simply be BOINC evening-out the historic resource share: if SETI has received the lion's share of the computer's time in the past, then BOINC will preferentially fetch from Einstein for a while to level the playing field.


If I thought that was all that was happening, I wouldn't worry about it. But it seems to take huge doses of Einstein at a time, run them preferentially because of their shorter deadlines, and then do none for a while.


If Einstein tasks are running preferentially - I assume you mean in 'high priority', or EDF - then may I respectfully suggest that your cache settings are too high for running two projects in tandem.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: RE: RE: RE: N9JFE

Quote:
Quote:
Quote:
Quote:
N9JFE wrote:
let its Seti tasks on board fall below the server limits.
What do you mean by that?

SETI has project-specific limits of "maximum 100 GPU tasks in progress" (and the same for CPU tasks, making a limit of 200 tasks in total).

This will simply be BOINC evening-out the historic resource share: if SETI has received the lion's share of the computer's time in the past, then BOINC will preferentially fetch from Einstein for a while to level the playing field.


If I thought that was all that was happening, I wouldn't worry about it. But it seems to take huge doses of Einstein at a time, run them preferentially because of their shorter deadlines, and then do none for a while.

If Einstein tasks are running preferentially - I assume you mean in 'high priority', or EDF - then may I respectfully suggest that your cache settings are too high for running two projects in tandem.


Hmm. I thought I had changed that, but I just checked and it was set for additional 10 days. I changed it to 6. I'll see if that does anything.

It's Boinc 6.10.60.

...

I just checked my other box, which seems to maintain a more balanced mix of tasks. On Boinc 6.10.58, it's also set for additional 10 days, but the "connect every" is set to 4 days, whereas the one that's moody is set to .25 days.

(Looking at that one (the one with the GT 630) now makes me wonder what the heck is going on with Seti this morning, but I'll check that out over on their boards.)

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.