I have 11 work units lined up, all of a sudden, my computer has preempted a work unit, starting on another.

In the past, it would complete a unit before starting on the next.

What would cause this problem ? I am only crunching SETI work units. Do I have a setting wrong ?

Thank you for any information.

---

If it feels one will not finish before it's due date, it will prempt other results, and do the one that it thinks it will not finish, then go back.

---

My movie https://vimeo.com/manage/videos/502242

If it feels one will not finish before it's due date, it will prempt other results, and do the one that it thinks it will not finish, then go back.

The unit it was working on, due date 10/05 was about half finished, then it started on a unit with a deadliine of 10/12, now it's working on the 10/05 unit again.

I'd think it would stay on the 10/05 deadline before moving on.

The unit it was working on, due date 10/05 was about half finished, then it started on a unit with a deadliine of 10/12, now it's working on the 10/05 unit again.

I'd think it would stay on the 10/05 deadline before moving on.

Not necessarily. The longer due date means that it take a longer time to crunch that unit. So, depending on the information that BOINC has (time machine is on, time it's allow to work, results duration factor, etc.) it could say that the longer unit will be late, so it will do it.

---

My movie https://vimeo.com/manage/videos/502242

If the next result that was supposed to be started by FIFO (which is how an individual project is run normally) would cause a result that is later in the queue but has an earlier deadline to be late, the earlier deadline result will be started. However, if the client detects that the result that is later in the queue will be completed on time, then the one earlier in the queue may be started. This behavior may be different depending on the version of the client. Then later when the scheduler notices that the earlier deadline result may be late, that one will be started again.

Since this is done on an hourly basis, not much work is lost.

This behavior only started at S@H since there have been results with varying deadlines and to completion estimates. If I recall correctly, 5.6 will prefer to crunch a result that has already been started as opposed to starting a new one, however, if it needs to run a project in EDF for a bit, it will do so. At that point, there are likely to be two (or possibly more) results started.

---

BOINC WIKI

I really wish preempting would not happen for 1 reason. Every work unit eats up about 60 MB or so of RAM. Right now on my main PC I noticed that 2 work units are preempted so with the current unit in progress, SETI is consuming over 180 MB of RAM - 120 of which is wasted by 2 preempted units (1 of them has been 0.04% processed - what a waste!!!).

It's not a huge issue on my main PC which has 1 GB of RAM but on my other PCs which have 512 MB, 384 MB, & 256 MB, this can be a major headache (especially on the 384 MB PC - it has dual CPUs so if each one has 1 preempted unit then over 240 MB are being used). Hopefully the preempted units are moved to virtual memory when necessary.

---

It's not a huge issue on my main PC which has 1 GB of RAM but on my other PCs which have 512 MB, 384 MB, & 256 MB, this can be a major headache (especially on the 384 MB PC - it has dual CPUs so if each one has 1 preempted unit then over 240 MB are being used). Hopefully the preempted units are moved to virtual memory when necessary.

Virtual memory is managed by the OS. If the OS sees that a page of memory is not being accessed as much as another, it will shuffle it off to disk.

If you don't want preemtped work in RAM, there is a setting that allows you to turn that off.

You'll find it under "your account" here on the SETI site (or any other project site where you participate).

---

Ok Ned, I'll bite, I looked and unless I'm blind I didn't see anything like that there. And I looked in both General preferences and SETI@home preferences.

General Prefs>Processor Usage>Leave Apps in Memory While Suspended: Default is NO

Alinator

Funny, My PCs are never suspended, But yet It does pre-empt WU's to work on other WU's and I'm only doing Seti.

Small cache, won't need to pre-empt often. Larger cache, will have to pre-empt more often.

---

Grant
Darwin NT

Ok Ned, I'll bite, I looked and unless I'm blind I didn't see anything like that there. And I looked in both General preferences and SETI@home preferences.

General Prefs>Processor Usage>Leave Apps in Memory While Suspended: Default is NO

Alinator

Funny, My PCs are never suspended, But yet It does pre-empt WU's to work on other WU's and I'm only doing Seti.

This is just depricated wording. It doesn't mean remove from memory if BOINC or a project is suspended, it means if the app isn't running remove from memory.

The single project pre-empting effect is due to the conservative scheduling "Fudge Factor" built into BOINC, in combination with the short deadline results, when they show up, and your cache size.

3.15 days is what mine is set at currently, 0.15 days was too small before the last outage.

Set it at one or less, pre-empting is unlikely to occur. More than a day or 2 & pre-empting will occur. It's normal, it's desirable. It's good that it does occur. It's not a problem.

---

Grant
Darwin NT