Some Examples of Why You Should Pay Attention to Your Boxes
2 Nov 2006 18:31:26 UTC
Topic 192013
(moderation:
Case 1: Why It's a Bad Idea to Set a Max Cache when Crunching Multiple Projects.
Pretty self explanatory.
Case 2: Why You Shouldn't Assume BOINC *Always* Knows What It's Doing.
For this machine, this represents roughly 33,950,000 seconds worth of work (or ~393 days). Deadlines might become an issue here. Not only that, but since the host is actually trying to work through them, as far as I can tell, this might lead to "orphaned" result in the DB as it reports expired results which have been purged. It would be interesting to see what the last few contact log entries for this host looked like.
Alinator
Some Examples of Why You Should Pay Attention to Your Boxes
)
Note for Case 2: The 33 MSecs might be a little harsh, lets call it 10-15 Msecs assuming it drew a mix of long and short, but you get the idea. ;-)
Aliantor
*lol* That's kinda
)
*lol* That's kinda informative. Do you have any information if the person representing case 1 has been hunted down by other crunchers waiting for validation of their WUs? ;-) Just kidding, but I know how annoying it can be if you usually finish WUs after a day or so and regularly get teamed up with a guy who takes five or more, so... 14 days... wow ^^ Maybe some people should really think again about their cache size.
What about this
)
What about this guy
http://einsteinathome.org/host/33556
I don't know what he is doing but this host seems only to download WU's instead of actually processing them. I know I'm waiting for him to finally turn in some results. :(
Yep, I'd sure like to know
)
Yep, I'd sure like to know what happened to cause BOINC to go insane like this. It looks like it was running fine until around 8 to 10 days ago, and would have had to draw a full compliment of 144 results a day for that time period get to this level (1689 results in progress).
At this point, just aborting most of them would take a while. :-)
Alinator
It definitely would ^^
)
It definitely would ^^ finally I've found someone whose list looks worse than my CPDN one with 251 results, 250 of them unprocessed... but in that case, it was a server bug and the first 250 WUs never really reached me, whereas Einstein servers seem to run fine, so I assume that guy really has all those 1600 WUs on his box... happy crunching with those :-D at least he can be sure his box won't be idle for the next years or so.
Well, the good news is if it
)
Well, the good news is if it doesn't start reporting results pretty quick, the host will start getting throttled back on new work sent in about 20 hours when the oldest ones start expiring.
The bad news is, there's a of lot hosts which are going be waiting for validation as this pretty hefty chunk gets reissued and run through the mill again.
Alinator
RE: The bad news is,
)
But the typical host will only be waiting for an additional two or three weeks....
Director, Einstein@Home
I just checked one machine
)
I just checked one machine where the person I'm paired with has 1040 wu's, of which he shares 112 with me, and none of those 112 have been returned. Doesn't look like they will be either.
RE: I just checked one
)
Someone has recently suggested a modification to the BOINC scheduler so that it will not send additional work to hosts that have too may workunits 'in progress'. I'll have a look -- this might help solve the problem.
Cheers,
Bruce
Director, Einstein@Home
RE: Someone has recently
)
The people doing the wu's need to pay attention to what their machines are doing and take action when necessary. There's always that handy little "No New Work" button if the cache becomes to large. With the almost total lack of downtime on Einstein I've never seen any reason to carry a large cache, but, I still check my boxes on a daily basis to make sure everything is OK.