Forum > Discussion Forum
Seti is down again
msattler:
--- Quote from: Frizz on 22 Oct 2010, 06:31:27 am ---
--- Quote from: msattler on 22 Oct 2010, 06:25:47 am ---OMFG.......if you think that is an answer to the project's problems.......
I better just stop now.......otherwise the kitties are just going to go berserk.
Grrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr.
--- End quote ---
Better stop drinking - and start reading. Properly. Nobody want's to stop optimized apps ... ::)
--- End quote ---
Who said I was drinking........I'll tell you when I an f$$$$ing drinking.........not nice of you to presume.
Miep:
... and I already have a migraine...
So Right first up:
Nobody suggested that only big users should be considerate of the current servers reduced capacities.
Nothing we can do about the crowd, that doesn't read forums etc. or doesn't even remember what that blue-yellow symbol in their taskbar is. And that is almost certainly the majority of the user base.
Those with eyes to read however might want to consider what they can collectively do to reduce server load.
Yes, Mark, we don't gain much if just the handfull of big ones take half a step back - it will only have an effect if everybody contributes. And of course NNT on an empty rig is not really an option [well unless you run several other projects and can switch to them to give SETI a breather]
What else? NNT and manually update once in a while. Set small cache until the worst has passed. Don't keep hitting the update button (especially not in the first half hour when everybody is homing in) I'm sure there's more tweaking options (i.e. spreading with backoffs) that can be used to 'take half a step back'.
As Richard has already pointed out, while the limits help to distribute work more evenly they increase serverload because the starved GPUs keep asking for more only to be told they already got some...
It would probably help to implement a 'you've reached a limit'-backoff, so there won't be essentially empty requests. Dripfeeding, while keeping the rigs at the limit, also doesn't help with serverload.
Effectively the 'report one, get one' is the same as if you were set to report after each task - and it makes a vast difference to serverload if you do 10x 1 or 1x 10.
So the best suggestion I can currently come up with is to intermittently set NNT: Wait till you have some work [well when the system is finally up to it] by letting boinc run it's usual backoff routine. Fill up a bit - maybe half a day worth, maybe to the limits - then go to NNT. After a few hours (i.e. while you still have some left, so you won't run dry but so that you will be asking for a reasonable amount) switch back on. fill up some, switch off. redo from start...
I've no idea how well that works, or if it is feasible for larger systems. It does require some babysitting, but it will help to reduce serverload. As to ghost creation I wouldn't think there's a different chance of a single request getting lost than of a bunch - statistically the single requests might result in less ghosts per system, but at the cost of increasing serverload.
So bottomline: If everybody could try to be a little less emotional and egoistic, we might be seeing some work before the new servers are in place instead of one crash after the other.
And as always, talking - sorry writing - to the wrong audience ::)
Raistmer:
Filling cache then go to disabled network has another good side:
You will accumulate lot of results to scan for elepsed times. Then you will have very nice picture of host true performance (for beta apps, copying whole project directory before switch network ON again will provide good backup for offline checking tasks that can produce invalid results online.
So, switching ON, loading cache, then switching OFF network completely - is the way I run my ATI box many weeks already.
Raistmer:
BTW, on my good old Athlon 64 I have 3 SETI MB tasks in download state now ... and absolutely idle CPU :(
Einstein backup refuse to ask for work. BOINC thinks that SETI tasks should be run instead (but they not downloaded and will not be downloaded any time soon).
Obviously it's BOINC bug. Anyone knows is it fixed already and in what BOINC version?
Jason G:
--- Quote from: Raistmer on 22 Oct 2010, 10:11:25 am ---BTW, on my good old Athlon 64 I have 3 SETI MB tasks in download state now ...
--- End quote ---
Ahhh, so you're the one blowing up the servers. ::)
--- Quote ---Obviously it's BOINC bug. Anyone knows is it fixed already and in what BOINC version?
--- End quote ---
Not sure which Boinc version it was introduced in, fairly recent I think.
[Edit:] http://boinc.berkeley.edu/trac/wiki/VersionHistory says 6.10.19-56 , so probably buggy I suppose. Will see if I can see anything obvious in my modded 6.10.58 this weekend.
[Later:] Check your debts, and also whether those 3 tasks in the download queue are blocking all requests for work (Shouldn't be :-\ any uploads ?).
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version