LHC@home

Hello, dear friends!
Problem.
Once LHC task(s) is (are) downloaded, the tasks from one of other BOINC projects begin Running in high priority and blocking LHC. That one of other is always the project with shortest deadline. But I was shocked, when a host, usualy running CPDN only, captured LHC tasks, and CPDN tasks began Running in high priority. This is absurdity!
Details.
1. No matter:
- how big are LHC tasks (for 20 minutes or 20 hours);
- what is Resource share between projects;
- which value is set for Additional work buffer or how much tasks have other projects.
2. It is important, which version of BOINC is installed:
- this effect appears, when BOINC manager is 6.6 or higher;
- there is no problems, when manager is 6.2 or lower.
P.S. This situation is never seen before...

---

Hello, dear friends!
Problem.
Once LHC task(s) is (are) downloaded, the tasks from one of other BOINC projects begin Running in high priority and blocking LHC. That one of other is always the project with shortest deadline. But I was shocked, when a host, usualy running CPDN only, captured LHC tasks, and CPDN tasks began Running in high priority. This is absurdity!
Details.
1. No matter:
- how big are LHC tasks (for 20 minutes or 20 hours);
- what is Resource share between projects;
- which value is set for Additional work buffer or how much tasks have other projects.
2. It is important, which version of BOINC is installed:
- this effect appears, when BOINC manager is 6.6 or higher;
- there is no problems, when manager is 6.2 or lower.
P.S. This situation is never seen before...

This is a new feature in Boinc 6.6 maybe also in 6.5
Now not the result with the first deadline will calculated, but in Boinc 6.6 the result with the most time problems
is calculated first.
Had this seen on my Boinc also, after a short time all goes back to normal work.

---

Matthias

Indeed, the scheduling behaviour has been changing a -lot- over the past releases. I believe the 6.10.x builds introduced a strict first-in-first-out scheduling mode, and since then there have been various attempts to fix resulting task preemption issues. I\'m not sure whether the latest fix made it into the latest build - 6.10.37 - the fix is on Branch but the release was never tagged.

The problem of a project going into high priority when new work from a different project is received is often caused by low Resource Share for the project in HP. CPDN is especially sensitive when it has low RS because of its long running WU.

---

Questions? Answers are in the BOINC Wiki.

Boinc V6.10.56 Recommended
WinXP C2D 2.1G 3GB

I\'m not sure whether the latest fix made it into the latest build - 6.10.37 - the fix is on Branch but the release was never tagged.

Ah, to answer my own question, no it wasn\'t. So I guess we\'ll have to wait for 6.10.38 before we can see if the change finally fixes things.

Well, 6.10.39 is now out, including the latest scheduler fix. You can find it here. Remember that this is an experimental build - though I don\'t anticipate any problems.

The new Version is 6.10.40 for Windows

---

Matthias

Well, 6.10.39 is now out, including the latest scheduler fix. You can find it here. Remember that this is an experimental build - though I don\\\'t anticipate any problems.

BOINC still shows 6.10.18 as the recommended (stable?) version.

Well, 6.10.39 is now out, including the latest scheduler fix. You can find it here. Remember that this is an experimental build - though I don\\\\\\\'t anticipate any problems.

BOINC still shows 6.10.18 as the recommended (stable?) version.

There was nothing really major wrong with .39 other than they missed a couple of bug fixes that should of made it into the release. Basically, an accounting error was made compiling it. They are now testing .43

---

Questions? Answers are in the BOINC Wiki.

Boinc V6.10.56 Recommended
WinXP C2D 2.1G 3GB

Thanks to all, who wrote replays here.
I upgraded a part of my machines to newest recommended BOINC version - 6.10.43. No difference, excluding some new commands for better control of GPU - AI of BOINC manager is still not able to calculate, how much is 2x2.
An example (actual to 6.10.43):
- Einstein has work for 2-3 days by deadline after 2 weeks;
- LHC task with estimated CPU time for 1 hour is downloaded;
- Einstein is going to panic mode immediately and not for hours - for days;
- LHC is absolutely blocked.
This situation is a true absurdity.
Second bad thing, actual to BOINC 6.10, is - if computer has enough work, BOINC manager is connecting to servers for reports and requests (no matter, which project, excluding CPDN) very rarely - for example, once in 1 day.
So, the question is - what to do?

---

Here are the rules followed by Boinc for reporting completed work as posted at Seti by JM7 who did much of the scheduler tweeking on V5.

BOINC returns work to a project at the first of:

1) 24 hours before deadline.
2) Connect Every X before deadline.
3) 24 hours after completion.
4) 60 seconds after the completion of upload if upload completes after any of 1 to 3.
5) On a work request to that project.
6) With any other reported work for that project.
7) On a trickle up to the project (CPDN only as far as I know).
8) On a trickle down request (no projects use this as far as I know).
9) On a server specified schedule (some project uses this, but not certain which).
10) On a BOINC Account Manager specified connection (only if using a BOINC Account Manager).
11) When the user pushes the update button.

---

Questions? Answers are in the BOINC Wiki.

Boinc V6.10.56 Recommended
WinXP C2D 2.1G 3GB

1) 24 hours before deadline.
2) Connect Every X before deadline.
3) 24 hours after completion.
4) 60 seconds after the completion of upload if upload completes after any of 1 to 3...

Yes-yes-yes. It would be great, if this would be a truth!
2) is not working for BOINC 6.10 (including newest 6.10.43). Das funktioniert nicht! No matter, which value is set for X.

Thank You for reply, Aurora.

I have seen the high priority behavior discussed.
I have seen it only on computers where I am running
more than three projects. I have found that

1.) I can set one of the other projects to not
get more work. When there are only three
projects have work left, the scheduler works
again.

2.) I can suspend one or more of the projects for
a while and the scheduler operates correctly
for the remaining projects.

You might consider only running three projects at a
per machine....

I have had that problem several times with Einstein so if it is stopping me from doing ANY LHC units I just suspend Einstein until the LHC are done and then turn Einstein back on and I have never finished them past the due date that way.

---

Volunteer Mad Scientist For Life