FLOP estimates for SETI@home workunits changed.

Message boards : News : FLOP estimates for SETI@home workunits changed.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Eric J Korpela
Volunteer moderator
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 15 Mar 05
Posts: 1547
Credit: 26,833,922
RAC: 1,354
United States
Message 44282 - Posted: 29 Oct 2012, 21:58:46 UTC

I've finally changed to work estimates to include autocorrelation. For the short term, the change means that the server and your client will overestimate run times until it has averaged the new estimates into your run time stats.

No action on your part is necessary, but let me know if you get totally outrageous run time estimates.
ID: 44282 · Report as offensive
Richard Haselgrove
Volunteer tester

Send message
Joined: 3 Jan 07
Posts: 1444
Credit: 3,265,669
RAC: 0
United Kingdom
Message 44283 - Posted: 29 Oct 2012, 22:15:06 UTC - in response to Message 44282.  

Presumably this will apply to newly-split work only.

Do we have any way of knowing how many of the 5,857 tasks ready to send are MB, and how many AP? That would help us to know when to start watching for the new estimates.
ID: 44283 · Report as offensive
Profile Eric J Korpela
Volunteer moderator
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 15 Mar 05
Posts: 1547
Credit: 26,833,922
RAC: 1,354
United States
Message 44284 - Posted: 29 Oct 2012, 23:05:41 UTC - in response to Message 44283.  

Most all of them are SETI@home results. I've been filling the SETI@home send queue up to 4096 then letting it drain to 1024 before starting again. It just finished a fill a couple hours ago.
ID: 44284 · Report as offensive
Josef W. Segur
Volunteer tester

Send message
Joined: 14 Oct 05
Posts: 1137
Credit: 1,848,733
RAC: 0
United States
Message 44285 - Posted: 31 Oct 2012, 2:47:32 UTC

For the ~0.42 AR of the tasks the estimate adjustment is just over a 14% increase, as specified in the code change and as seen on my host 10490. For that 1.4 GHz Pentium-M the estimate change was from 8:36 to 9:49. That's hr:min, but a GPU which is 60 times faster would have similar min:sec change.

With duration correction factor (DCF) being much quicker to react than the server averages, any host which has done 10 or more of the tasks with new rsc_fpops_est probably already has displayed estimates close to real run times. But because the servers ignore DCF the amount of work sent for a request will be somewhat less, perhaps leading to an additional request. Because the servers on average send an extra 1/2 task beyond the request, that effect may be hard to see.
                                                                    Joe
ID: 44285 · Report as offensive
MindzX
Volunteer tester

Send message
Joined: 19 Oct 11
Posts: 1
Credit: 45,187
RAC: 0
United States
Message 44319 - Posted: 14 Nov 2012, 6:32:13 UTC

if you get totally outrageous run time estimates.


what constitutes this?
ID: 44319 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 29 May 06
Posts: 1037
Credit: 8,439,427
RAC: 0
United Kingdom
Message 44320 - Posted: 14 Nov 2012, 13:33:29 UTC - in response to Message 44319.  

if you get totally outrageous run time estimates.


what constitutes this?

Your host hasn't completed it's 10 non-outlier tasks yet for the CPU MBv7 6.98 app, or the OpenCL Astropulse apps, once it does, the estimates will be close,

Claggy
ID: 44320 · Report as offensive

Message boards : News : FLOP estimates for SETI@home workunits changed.


 
©2019 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.