Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go »
No member browsing this thread
Thread Status: Active
Total posts in this thread: 26
Posts: 26   Pages: 3   [ Previous Page | 1 2 3 ]
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 5447 times and has 25 replies Next Thread
SekeRob
Master Cruncher
Joined: Jan 7, 2013
Post Count: 2741
Status: Offline
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

... and indeed minutes later 8 arrive [in bits an pieces].
[Feb 27, 2016 6:10:04 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Mumak
Senior Cruncher
Joined: Dec 7, 2012
Post Count: 477
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

Odd is how the estimated task runtime works in this case. When such short tasks start to appear their estimated time is same as the longer ones. Later, the time starts dropping, but when the supply of short WUs stops, the estimated runtime of the longer tasks is that of the short ones, while the shorter ones are estimated to take longer. This also causes a huge overload of longer tasks downloaded.
Here a sample what happened after the short ones stopped and normal ones started flowing in:

OET1_0001862_xSDGP-S_rig_37070_0: est runtime 00:17:01
OET1_0001846_xSDGP-FW_rig_71171_2: est runtime 00:56:16

That's the exact opposite of reality.

Both tasks have almost same est. computation size -FW: 10038 GFLOPs , -S: 10168 GFLOPs
How does this estimation work, is only the client calculating this ?
----------------------------------------

----------------------------------------
[Edit 1 times, last edit by Mumak at Feb 27, 2016 8:10:10 PM]
[Feb 27, 2016 8:04:55 PM]   Link   Report threatening or abusive post: please login first  Go to top 
SekeRob
Master Cruncher
Joined: Jan 7, 2013
Post Count: 2741
Status: Offline
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

We're done with shorts it seems... got 1862-1869 and the occasional 1872, meaning soon we'll be making the batch jump forward again, to then step back and do [almost] all those in-between. Runtimes now in the 1.5-2 hours range on the 3770K at 3Ghz.

Mumak how it functions has been discussed frequently... it works on a feedback loop, the projected FPOPS for new tasks fed off the running average of work being returned i.e. always trailing reality.
[Feb 28, 2016 2:09:32 PM]   Link   Report threatening or abusive post: please login first  Go to top 
SekeRob
Master Cruncher
Joined: Jan 7, 2013
Post Count: 2741
Status: Offline
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

The whippets are were back on the track again... me 3770K is was racing through tasks from batches 1852-1854. Depleted those and 1860 is 'normalish' once more, and since there's serious fluctuating lapse-rate in the work generator to feeder nozzle, got the full 280 max of those sent at 19:36 minutes, but really taking 1:10 hours... that's 2 days before there will be any backfill again. nerd
----------------------------------------
[Edit 1 times, last edit by SekeRob* at Mar 13, 2016 5:19:33 PM]
[Mar 13, 2016 5:18:05 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Mumak
Senior Cruncher
Joined: Dec 7, 2012
Post Count: 477
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

Yep, that's the problem I'm experiencing too. Also the credit claimed acts similar - once the short tasks start it's higher, then upon return of normal tasks these get a lower credit than usual.
BTW, originally each short task supply lasted ~2-2.5 days, the recent series lasted however only for 1 day. So it seems the supply is decreasing.
----------------------------------------

[Mar 13, 2016 7:05:22 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Speedy51
Veteran Cruncher
New Zealand
Joined: Nov 4, 2005
Post Count: 1326
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Very Short Tasks

OET1_ 0001894_ xZAGP-FW_ rig_ 46544_ 0-- ran for 6.6 minutes on a I 7 5960 X at stock 3.0 Windows 10 professional 64 bit
----------------------------------------

----------------------------------------
[Edit 1 times, last edit by Speedy51 at Mar 28, 2016 3:36:44 AM]
[Mar 28, 2016 3:35:09 AM]   Link   Report threatening or abusive post: please login first  Go to top 
Posts: 26   Pages: 3   [ Previous Page | 1 2 3 ]
[ Jump to Last Post ]
Post new Thread