News:

If you are a member of the Team on BOINC you still need to register on this forum to see the member posts.  The posts available for visitors are not posted to much by members.
 Remember to answer the questions when Registering and also you must be a active member of Team BOINC@AUSTRALIA on BOINC.

Main Menu

Host Limit

Started by Furlozza, December 12, 2009, 01:08:24 PM

Previous topic - Next topic

Furlozza

 :pcwhack:

So, just checked xterminGnator to see how he was doing at Yoyo, and looked in at the messages and was surprised to see that I was being advised that xtermiGnator had "reached per host limit of 50 tasks".

I suppose that is one way of spreading the work around.

kashi

I hadn't experienced the per host quota of 50 tasks. Shouldn't stop you though because as soon as you have completed and returned some you should be able to download some more up to the maximum of 50 in your cache at once.

I do know the "Maximum daily WU quota per CPU" of 500 a day was set to try and prevent anonymous cherrypickers from downloading many thousands of tasks per CPU and just doing the ones that have a higher credit rate. ecm tasks are very variable in length, some of them are very short and run for only a few hundred seconds. Some ecm tasks give a much higher credit rate than others, I think it is the longer running ones* that are being assigned recently and some people were finding out which ones were the high credit ones and just crunching them and aborting all the lower credit ones. Possibly applies more to the 64-bit application which is probably much more efficient at ecm kind of tasks. Not a new concept for BOINC, I had noticed anonymous cherrypickers at Einstein and ABC in the past.

* Yep, just checked, it's the long ones that are big payers, completed about 6 of these so far, all about the same runtime and identical credit:

Nehalem @ 1.6GHz, Windows 7 64
ecm 64-bit Windows application.
20,347.63 seconds 253.50 credits

Thanks for letting me know about this 50 WU limit, I just filled up my cache with 50 of these biggies.

Furlozza

Hey, that sounds like cherry picking to me *WG*

Also noticed that Evo can run more than supposed time as can Muon. I know that they want you to put them into memory as they don't have (in at least one case, and can't recall which one) checkpoints......... sooooooo if you stop the wu for any reason (I had eight AQUA minis that were approaching wipe out time and so suspended Yoyo and WCG whilst they completed) there is a good chance that the wu will start from square one, but NOT reset the time therefore making you think that it is going .... well too far.

And yet there are occurrences where either of them will run OT and yet still be running the same program without having gone back to square one.  So I guess it will be a matter of watch, wait and see, unless there is a definitive answer out there somewhere.

kashi

No not me, cherrypickers abort WUs that have a lower credit rate. I just increased my cache to a few days size at an opportune time.  >:D

Normally prefer a shorter cache but ecm tasks do not have a quorum so I won't be holding anyone up. I'll be running both ecm and Muon, once I get my little ecm paperclip looking badge.

It's Evolution tasks that doesn't have checkpoints, that's why I can't run them.

As for how often the different subprojects checkpoint and how they display percentage completed and Elapsed time after a restart I found a detailed explanation somewhere in the forum yesterday but I've looked and looked and can't find it again for you.

I know the ecm tasks I am currently doing have progress percentage steps every 20%, but I don't know how often they checkpoint.

Muon I don't remember and cruncher OGR tasks had a poor rate of credit on my computer and one task idled for hours and the completed percentage went up to 6,300%, so it had obviously crashed and I aborted it after it had run for 14 hours. So I am not doing any more of those.

I can understand you not wanting to leave tasks in memory if you are sharing resources with other projects, some of the ecm tasks use large amounts of memory at various times in their processing cycle. I am currently running only 6 cores on ecm and no other CPU projects to minimise contention issues because of this.

kashi

Not sure if this is up to date but found these details here: www.rechenkraft.net/wiki/index.php?title=Yoyo%40home_en

Cruncher ogr

Checkpoints - nearly every 15 minutes checkpoints are written
Progress indicator - after processing of 1 of 7 ogr the indicator is increased by 14,2%, but this may need up to 5h
Credits -   based on processed GigaNodes


evolution@home

Checkpoints - not available, but with "keep in memory" also suspend is possible
Progress indicator - based on a rough runtime estimation the progress is shown. Since it is only an estimation, the work unit may already finish at 40% or at 150%.
Credits - based on calculated GigaIndividuals

Muon

Checkpoints - every 4 minutes
Progress indicator- will be increased every 33.3%. If a simulation produces a very good result it will rerun 4 times to verify it. In this case the progress indicator increases by 6.6%.
Credits - based on iterations

ecm

Checkpoints - every 10 minutes
Progress indicator - yes, every 20%
Credits - based on complexity

ecm is taking about an hour between checkpoints on my computer on the current long tasks, so any of the checkpoint times shown above may be very inaccurate in some cases.

Furlozza

Uhuh.

That is the bit I saw as well, and can confirm that it is Evo that runs over the said time.... but still comes in around the 35-45 cr/hr rate.

The muons I have done have come in below 20 cr/wu, but that works out at (from memory) about 30-40 cr/hr.

Either way you slice them, still better than SETI *grin*