Posts by dbrinza

21) Message boards : Number crunching : Is boinc.setiatwork.com dead? (Message 230521)
Posted 13 Jan 2006 by Profile dbrinza
Post:
I've tried to check my stats on boinc.setiatwork.com a few times today with no response. Anybody know what's up??
22) Message boards : Number crunching : Did this WU/result slip through a crack? (Message 230044)
Posted 12 Jan 2006 by Profile dbrinza
Post:
There have been cases where a result is returned and for one reason or another it never seems to be validated. It will be interesting to see if this one becomes one of them. In theory, when the transitioner comes back around it should pick yours up and allow it to be validated.

There may be a minor bug in the logic somewhere there ...

So, does the transitioner come back around ~14 days after the last of the 4 results were sent for processing?

Based on what I understand from BOINC-Wiki (excellent resource!), I'm wondering if somehow the WU got dropped in the "waiting for transition queue" after my result was returned. There was alot going on in Berkeley, including multiple reboots, changing the result filenames, rebuilding some of the backend processes, database merge, dropped connections, ... more fun than anyone deserves.

Maybe my result gets a second chance when it's WU is transitioned for the file deleter?? I'll be looking.
23) Message boards : Number crunching : Did this WU/result slip through a crack? (Message 228605)
Posted 9 Jan 2006 by Profile dbrinza
Post:

The first three were too far apart to allow the validation "box" to fit over the top. The 4th comes in and it is in the "middle" and bing!, now all 4 fit within the box ...



Right, Paul, that is exactly what I saw when no consensus was reached with the first three results, but when the fourth results arrived, all results were granted credit for that WU.

The WU at issue in this thread had validated the first three results and granted credit. My result was returned later that same day (and only 25 hours after it was sent to my computer), but the credit is still pending. There is no mention of client error, or other bad outcome for my result.

Here's the WU in question:

http://setiathome.berkeley.edu/workunit.php?wuid=44817228
24) Message boards : Number crunching : Did this WU/result slip through a crack? (Message 228425)
Posted 9 Jan 2006 by Profile dbrinza
Post:

If for whatever reason its NOT validated, it will just expire and someone else will pick it up. Why it is not validated, I don't know though.

A few weeks ago, I had a WU that not not validate after the third result was returned. The log for the result unit showed the validation state as "Checked, but no consensus yet". See:

http://setiathome.berkeley.edu/forum_thread.php?id=25977

Ultimately, after the fourth result was returned, that WU was fully validated and all of the users returning results for the WU were granted credit.

I'm seeing something I haven't seen before with the WU that is the subject of this thread. Any thoughts from Paul or Matt??
25) Message boards : Number crunching : Did this WU/result slip through a crack? (Message 228314)
Posted 9 Jan 2006 by Profile dbrinza
Post:
The following WU has not completed validation even though all four results were returned by Jan 1:

http://setiathome.berkeley.edu/workunit.php?wuid=44817228


Credit has been granted for the three other results submitted earlier that day, but my result is "stuck" in the initial validation state. The WU was distributed on Dec. 31 so there are still a few days before the 14-day expiration hits.
26) Message boards : Number crunching : WHERE DID THE CREDITS GO? (Message 227811)
Posted 8 Jan 2006 by Profile dbrinza
Post:
I do undestand the difference in "earning" credits/WU between classic and Boinc, and that those stats are there. The question is, I had over 100 WUs listed as results already completed--that has now decreased to 46. I'm not questioning the credits earned at all. Thank you for helping!!!


So, you realize that after your result and the three others for a WU are validated (to determine credit awarded) and assimilated, that the WU is deleted from the database. Your credit accumulates and the science master database receives the "canonical result" for the WU. BOINC seti has been struggling with deleting the WU and results for some time, but now appears to have almost caught up. BTW, my WU results went from 106 to 45 in the past few days as well...
27) Message boards : Number crunching : No. Results per WU (Message 226529)
Posted 5 Jan 2006 by Profile dbrinza
Post:
That changing ratio most likely is just a artifact of the deletion process. A WU would be ready to delete once credit is granted, but not deleted then. I would expect the results files related to a specific WU have to all be deleted first before that WU can be actually deleted from the BOINC data base. The results deletion has higher priority and runs well ahead of the WU deletion process, so as it winds down you would naturally see that ratio climb due to this lag between the two processes. The ratio should go to a very high number, if you could track it, at the end when the results deleter runs out of work to do, and the WU deleter is still playing catchup with maybe as much as tens of thousands of WU to finish removing.


The above argument would apply if the ratio in question was WU/Results, but we're seeing the inverse (Results/WU) ratio increase. This implies that there are perhaps "orphaned" results (as suggested below) that are not getting deleted as the WUs are.
28) Message boards : Number crunching : No. Results per WU (Message 226492)
Posted 5 Jan 2006 by Profile dbrinza
Post:
I don't think you can assume that the deleter is going through finding a Work Unit for deletion then finding the related results and deleting them too.



I guess I did assume this sort of linkage, since WUs are validated and assimilated before they can be deleted. If the results are deleted separately, it seems there must be some look-up to indicate that it's associated WU had been validated/assimilated and is ready for deletion. I don't know enough of what's "under the hood" to speculate why the ratio has grown...
29) Message boards : Number crunching : No. Results per WU (Message 226470)
Posted 5 Jan 2006 by Profile dbrinza
Post:
Interesting observation in the current database purge: the ratio of #Results remaining to #WU has grown to over 7. Check out the plot I generated from "Scarecrow's" sahstats Data Table page:



I would expect this ratio to be slightly higher than 4 (client processing and download errors causing WUs to be sent to additional users), but it has grown over twice this value.
30) Message boards : Number crunching : Recent WU not validated after 3 results in (Message 220963)
Posted 24 Dec 2005 by Profile dbrinza
Post:
How interesting, I would have thought the computational alogorithms (FFTs and the like)to be quite deterministic. Is there some sort of random number based filtering in the "smoothing" process?

Almost ... :)

Look around for the term "Floating Point Number" to find threads where we talk about the accuracy problems with floating point.

LHC@Home does not have a Macintosh client at all. Not that the G4/G5 have bad floating point processors, just that they are sufficiently different, that combined with the compilers available the project was not able to get results that compared.

I have a very simplified illustration in the Wiki under Validation Process ...

The Wiki is your friend ... :)

Oh, and I talk a little bit about floating point there too as it applies to what we do ...


Paul:

Thanks for your response. I've poked around a bit as you've suggested and found very interesting threads discussing accuracy of floating point processors, whether sending out 4 WUs for processing really makes sense, etc.

In responding to "Crunc3r"'s post:
P.S. Even crunching the same wu with the standard setiathome client twice, gives slightly different results.
I wasn't talking about the (in)accuracy of floating point computations and differences in results encountered when processing on different platforms. I thought he was inferring that different results can be obtained by processing the same unit twice with the same application version on the same platform.

In my mind, the only way different results can be obtained for that case is if there is some time-varying parameter involved in the WU processing algorithms. In the absence of random-number generation, I suppose an overclocked machine just on the hairy edge might produce different results when chomping on the same WU twice if bits are flaky in the machine.

Or more simply, maybe I misunderstood Crunch3r's P.S. ... :)
31) Message boards : Number crunching : Recent WU not validated after 3 results in (Message 220363)
Posted 23 Dec 2005 by Profile dbrinza
Post:
It will not be rejected and you get your credit if the results are within acceptable range.

P.S. Even crunching the same wu with the standard setiathome client twice, gives slightly different results.



How interesting, I would have thought the computational alogorithms (FFTs and the like)to be quite deterministic. Is there some sort of random number based filtering in the "smoothing" process?

Edited comment: While I was posting the above, Tets' response appeared. I defer to the "heavy hitters" in this discussion...
32) Message boards : Number crunching : Optimized software and my results with a pentium m (Message 220358)
Posted 23 Dec 2005 by Profile dbrinza
Post:
Did the benchmarks from the optimized client show an increase in MIPS compared to the previous results? (Claimed credit involves the product of the Whetstone+Dhrystone with the CPU time).
33) Message boards : Number crunching : Recent WU not validated after 3 results in (Message 220354)
Posted 23 Dec 2005 by Profile dbrinza
Post:
So, will an "outlier" result be rejected (and granted zero credit) after the fourth result is reported and checked by the validator?
34) Message boards : Number crunching : Recent WU not validated after 3 results in (Message 220349)
Posted 23 Dec 2005 by Profile dbrinza
Post:
In the log of the result ID for each WU are logged "Checked, but no consensus yet". That means that three results has been returned, but the results differs to much. For example one or more are computed incorrect.

Maybe arithmetic or truncation errors occurred in the optimized application?
Thus far, I hadn't seen this happen with any other of my WUs.
35) Message boards : Number crunching : Recent WU not validated after 3 results in (Message 220342)
Posted 23 Dec 2005 by Profile dbrinza
Post:
I haven't seen this before and haven't found this in the message boards...

Here's a WU that has had three successful results returned (the third was reported over 8 hours ago), but hasn't been validated:

http://setiathome.berkeley.edu/workunit.php?wuid=42584692

The validators are up and running and the waiting for validation queue is near zero.

Did this WU slip through a crack??
36) Message boards : Number crunching : Oh, oh! Return of the download problem?? (Message 216916)
Posted 17 Dec 2005 by Profile dbrinza
Post:
The current status is posted on the home page, not in the Technical News page.

See the December 17, 2005 entry under "News" on:

SETI@home

Any guesses how long it'll take to fix the COgent problem? (The suggestion that we donate CPU cycles to other projects doesn't bode well for a rapid recovery)...


Oops! I just saw Matt's post below ... in a few days it'll be clear sailing again!!
37) Message boards : Number crunching : Oh, oh! Return of the download problem?? (Message 216913)
Posted 17 Dec 2005 by Profile dbrinza
Post:
The current status is posted on the home page, not in the Technical News page.

See the December 17, 2005 entry under "News" on:

SETI@home

Any guesses how long it'll take to fix the COgent problem? (The suggestion that we donate CPU cycles to other projects doesn't bode well for a rapid recovery)...
38) Message boards : Number crunching : User of the day. (Message 216898)
Posted 17 Dec 2005 by Profile dbrinza
Post:
WoW. I thought only people who have been with the project got user of the day. Thanks everyone :) It really is a pleasure :)


Well, as it turns out, Matt Lebofsky (a big-time SETI developer at Berkeley) got selected as "User of the Day". Check out his blog entry on August 2, 2005 regarding the "honor" of this recognition:

Matt Lebofsky: Slog

BTW - Congratulations, it's really cool to get picked. Maybe you should go buy a lotto ticket, since it's your lucky day!!
39) Message boards : Number crunching : Oh, oh! Return of the download problem?? (Message 216848)
Posted 17 Dec 2005 by Profile dbrinza
Post:
I'm getting the following messages:

12/17/2005 2:20:14 PM||Couldn't connect to hostname [setiboincdata.ssl.berkeley.edu]
12/17/2005 2:20:14 PM|SETI@home|Temporarily failed download of 13fe05aa.24442.6177.598568.175: system I/O
12/17/2005 2:20:14 PM|SETI@home|Backing off 1 minutes and 26 seconds on download of file 13fe05aa.24442.6177.598568.175

Seems like we just recovered from this problem. Are we having a relapse??
40) Message boards : Number crunching : Difference in computing (Message 216842)
Posted 17 Dec 2005 by Profile dbrinza
Post:
I installed optimized program on my pc, seems like computing time was cut in half, but now I lost visual graph, (iow, Show Graphics button is grey), is that normal ?

Yep, that's normal. The graphics slows the workunit numerical processing, so it's not supported in the optimized application.


Previous 20 · Next 20


 
©2024 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.