Apologies in advance if this is in another thread or in the wrong place.

I currently have an E2180 which is shared 50/50 on 2 different projects. One project is very easy on L2 cache requirements and the other extremely demanding. Best efficiency is achieved with a different project on each core of the CPU.

Recently the project, which is most demanding, has been starting a second WU and attempting to run both at the same time. This completely swamps my shared L2 cache and results in both WUs running at less than half normal speed and of course no work being done on the other project.

Is this situation preventable? Have I missed some settings? My only option atm is to disable downloads from the offending project and suspend one of the WUs. Then resume the WU after the other has completed. That works fine if I'm there to supervise but an automated method would be preferable.

Thanks in advance.
Ozylynx (Keith)

Thanks again Seksy.

Great to hear from you again and I'm pleased to see you're still being as helpful as ever.
btw. I finally worked out that my weird score discrepancies on WCG were due to L2 overflow and resultant cache swapping, on the older machines.
As always, I have what I think is a solution to that. It will wait for another time and thread though. lol.

[edit:] New idea. Is there a method of editing the machine ID within BOINC (windows)? Reason for asking; If I install a second copy of the BOINC client on a different partition and set both to run on 1 CPU core, I may be able to give each project 100% on 1 core each. There may be other problems, but the computer ID and stats would be the first that comes to mind. Has anybody tried this?[End Edit]

Happy trails.
ozylynx.

Affinity control might also be a problem as each client would not be aware of the other.

Seksy me

That's exactly what I was hoping for. I don't want affinity and if the CPU sees 2 different tasks it would, I think, assign a core to each internally. I see the 2 instances of BOINC being unaware of each other, part of the reason for giving each a seperate ID, as the main object of the exercise. The 2 cores then become 2 completely different computers as far as the project is concerned.

I used to do something very similar with HT machines on F@H simply by placing each instance of the client in different folders. The problem I forsee here is that BOINC client control is ultimately external and without having different names, would doubtless confuse the system.

Unfortunately, I have no knowledge of VM, so it's back to the drawing board.

Cheers

Affinity control might also be a problem as each client would not be aware of the other. Don't know VM to well and if it could be assigned to a particular core.

BOINC doesn't set affinity so I don't know what problem you're talking about.

---

Unfortunately, I have no knowledge of VM, so it's back to the drawing board.

OK, there are some already finished drawings to start with: a half year ago Dagorath was developing an add-on for Boinc (named SLM), which was, in your case, capable to limit your project A to at most one running WU and simultaneously limit your project B to also just one running WU. You may contact him for help.

In an ideal case, Boinc core client should be capable of this. In the past, trux has been incorporating such features in his Calibrating BOINC core client, unfortunately he went silent (although he still seems to be crunching).

Peter

Thanks Pepo. I'll see how they went. Sounds like just what I'm looking for.

Keith.

The up to date variants of Trux clients are available at Crunch3r's BOINC page.

The up to date variants of Trux clients are available at Crunch3r's BOINC page.

Crunch3r's Red Pill client should feature 'cpu affinity' and 'delayed return results immediately'. But he never talked about any of trux' additional features, and there were a plenty of them around :-)

I suppose Crunch3r's affinity is done automatically (start app on core 'n' and keep it there) and does not prevent running multiple concurrent WUs from the same project.

Peter