Weekly DC Stats - 07MAR2021

[Endgame124]

2 boinc instances is easiest on 2 separate hosts. Just setup another PC for LHC

 

[Markfw]

2 boinc instances is easiest on 2 separate hosts. Just setup another PC for LHC

Use Rosetta on the one that has the most memory.

Edit: Rosetta can up 1 gig per thread, sometimes up to 1 1/2 or 2. Hence the memory recommendation. I have at least one gig per thread on every box.

 

[voodoo5_6k]

Oh I see, what a daft system! I wish it would focus more on what's currently being crunched, any idea how long it would take to stabilise?
Atm I've turned the tap off for LHC in disgust, but if it's getting close to levelling out I could turn it back on again without Rosetta being totally excluded!?

Based on StefanR5R's explanation I'm currently testing this on my main rig. I chose WCG as the second project. My goal is to, and I assume that your goal is pretty much the same, keep the lion share of the host's resources for Rosetta, but give the second project a little room to breathe. However, the scheduler will try to bring the second project up to speed, and therefore will prefer it for a (long) while. Now knowing, that the scheduler is trying to achieve the resource share over the long term, based on recent estimated credit, and taking into account that Rosetta has been the one and only project on this host, I set a very low value (1%) for the WCG resource share (in the device settings on the WCG site), and a very high one (99%) for Rosetta (in the project settings on the Rosetta site). With this 99:1 ratio, my host is still running mostly Rosetta tasks, but will occasionally insert or continue to work on a few WCG tasks. Overnight, it completed 17 WCG tasks (which are short with ca. 2h each, compared to the Rosetta standard tasks with 8h). Now comparing my overnight Rosetta output with the usual output verifies that it hasn't been reduced too much. I'll be further looking into this over the next few days or even weeks, and try to find out what resource share will give me the best short term credit distribution.

Maybe tackling it this way will get you up and running in LHC without letting Rosetta suffer badly (because that's obviously how the scheduler is supposed to work- thanks again for your patient explanations, @StefanR5R ). And over time you can increment the LHC resource share until you reach the credit distribution you envision. It is of course not like assigning threads, for that the suggested separate BOINC instances might be the best way.

 

[Assimilator1]

I still think it's a daft system , they should at least give the option of thread division rather than RAC, i would think it would take several weeks for that to settle down?

On a separate note, I've just been distracted by a very loud pop and a flash from my laptop which was on charge!! No smoke came from it, the battery pack is cool and so was the charger (which it's power LED stayed on for several seconds as usual after I pulled the mains power), but their was definitely a smell of burnt electrics!
I quickly tried starting the laptop and it booted up just fine! lol (only let it run to the post screen as I need it for diagnostics on my car tomorrow, and don't want to drain the battery if I can't charge it anymore).
I don't get it, would could have blown?? m/brd cap??

 

[voodoo5_6k]

I still think it's a daft system , they should at least give the option of thread division rather than RAC, i would think it would take several weeks for that to settle down?

Yeah, it is not that I'd like it in particular, but I can arrange myself with it I would also prefer an explicit core/thread based distribution option, but as I'm not a developer of that application, I'll take it as-is, as long as it works.

I wouldn't say it takes weeks. At least not on a PC like mine, that has very few cores/threads and is therefore in a completely different ballpark compared to e.g. Markfw's. Based on my "experiment" with initially 99:1, and since yesterday afternoon 98:2, it has already almost settled. Your Rosetta output is roughly the same as mine, so I'd guess it wouldn't take your host much longer than mine.

On a separate note, I've just been distracted by a very loud pop and a flash from my laptop which was on charge!! No smoke came from it, the battery pack is cool and so was the charger (which it's power LED stayed on for several seconds as usual after I pulled the mains power), but their was definitely a smell of burnt electrics!
I quickly tried starting the laptop and it booted up just fine! lol (only let it run to the post screen as I need it for diagnostics on my car tomorrow, and don't want to drain the battery if I can't charge it anymore).
I don't get it, would could have blown?? m/brd cap??

Oh boy, you really have like a winning streak recently? Did you see where the "flash" originated? Otherwise, it is hard to say what might have died (or was severely wounded), without physical examination, i.e. opening the case and inspecting the electronics. In case the laptop works normally, it might have been the charger, but that's just a guess of course.

 

[Assimilator1]

Bit hard to say as it was in the corner of my eye, but it seemed to come from the back right of the laptop, the charger was at the front left of it, the charging port at the back left.

BOINC has stopped crunching the remaining few LHC WUs (I'd turned the tap off) and is allowing Rosetta to run again, so I've allowed LHC to grab WUs again. I'll see how it goes, if it hogs crunching time again I'll shut the tap off again semi-permanently.

Btw, re other peoples suggestion of running a 2nd PC, I'm not going to run one permanently.

 

[voodoo5_6k]

Bit hard to say as it was in the corner of my eye, but it seemed to come from the back right of the laptop, the charger was at the front left of it, the charging port at the back left.
[...]

Any news on the laptop? Is it still working normally? Maybe have a look at the hardware maintenance manual to check what's in that back right area without disassembling the unit (I have no idea what model you have). But I'd say it may not be a bad idea to preventively back up the data on it, just in case.

A quick follow-up, as I said I would look at this during the next few days or weeks. The BOINC scheduler has definitely settled. I changed the resource share from 99:1 to 98:2 on the second day, and to 95:5 on the fourth day. Each change was made after seeing the statistics curve slope decreasing significantly for the new project (WCG). Now, the seventh day is running, still at 95:5. So, on my system with relatively few cores, and giving the new project a tiny resource share with subsequent incrementations, it took the scheduler roughly a day to settle after each change. My intention was to not impact the output of the primary project too much, and that did work (as demonstrated by the Rosetta statistics graph).

 

[voodoo5_6k]

And finally, because I just couldn't let go, I ditched everything and now use BoincTasks

I have setup two instances of BOINC, named them accordingly and separated the projects that way. The first instance has 13 threads (=81.25%) and runs Rosetta, the second instance has 1 thread (6.25%) and runs WCG. The remaining 2 threads are for feeding the GPU in F@H. Both project's preferences are back to 100%. After a little messing with the client_state.xml and the project related files, everything works great. Even the WCG history is now on the new instance, just like I wanted. Cool tool It would be even cooler if it could start & end all BOINC instances on the local host, but it only starts/ends one (at least I haven't found a way to it yet). But anyhow, I like it!



Thanks @StefanR5R for the great How-To (How To - Multiple boinc clients on the same computer | AnandTech Forums: Technology, Hardware, Software, and Deals ).
(Maybe this could be a sticky so it can easier be found for new members?)

 

[StefanR5R]

a little messing with the client_state.xml

Be cautious with this. If both client instances happen to have the same CPID (cross project identifier), then take care that only one of the instances talks to a given project server. Safer is to have a unique CPID in each instance.

Maybe this could be a sticky so it can easier be found for new members?

Or we could put a copy at teamanandtech.org.

 

[Assimilator1]

Any news on the laptop? Is it still working normally? Maybe have a look at the hardware maintenance manual to check what's in that back right area without disassembling the unit (I have no idea what model you have). But I'd say it may not be a bad idea to preventively back up the data on it, just in case.

Seems to work just fine! lol
I'm not going to know for certain what blew without taking it apart, and I'm not going to bother with that.
Only data on there that I want is BMW diagnostic software which I have on DVD, so no worries there. It's only used for diagnostics.