Getting many invalids with FGRP2 over the last few weeks.
13 Sep 2013 15:53:54 UTC
Topic 197189
(moderation:
I´m getting lots of invalids lately, 6 in the past few weeks; never seen any before, so it raises some concerns. So far, they were all on an iMac, all FRGP2... The first ones were on v1.10, so I supposed this was the little mishap due to different compilers in use between Windows / Linux and OSX. A pity, as each times it´s 15 hours down the drain. However, last wednesday, I also got two invalids on FGRP2 v1.12! Something wrong with my Imac, time for panic? The problem only shows itself with Einstein as far as I am aware, but you never know...
See this small list, what I still can see are workunits 172730899 (v1.10), 173614624 and 173104915 (both v1.12). These are invalids for host 4687398.
Today though, I am a little bit relieved: for the very first time, I had an invalid on another machine, but that one is a very old Windows XP sp3 PC, host 7301083 and wu 173120284.
I wonder, has something else perhaps been changed, other than the new version 1.12? Something done over the past few weeks? Maybe with the work generator? Should I check something myself? I´m puzzled by the mishap, not to mention this irritation building up when seeing, as of now, 75 hours of crunching (on the Mac) and 46 (on the PC) going south, without me being part of it; the more so as they´re initially set to ¨Success¨. Somehow I wish it wasn´t bundled under these circumstances, hoping that odds are in favor of only a single unit being trashed when it so happens.
Getting many invalids with FGRP2 over the last few weeks.
)
I got a lot of these as well, at least 10 so far. Usualy when validating with v1.12 against two v1.10, but a couple of older 1.10 tasks failed against two 1.12.
Hopefully this will go away when everyone has 1.12 because wasting 24h+ of work isn't fun.
Got another one today:
)
Got another one today: http://einsteinathome.org/task/400616145
FGRP2 again. This one is not suffering version mismatch, all three are on v1.12, but yet again 15 hours down the drain. A pity, everything was going well until about a month ago, just before the latest version introduction. Seems to me it solved nothing, if I remember correctly I indeed had one invalid just in the week before update, so on v1.10, but a lot more now. It´s slowly becoming pointless and too painful to insist, sadly enough.
Hi! We are currently
)
Hi!
We are currently working on improving the cross-validation performance. We are dealing with time-series data of single photon arrival times where the time must be processed accurately down to micro-seconds precision, not trivial.
Occasional mismatches between results of the older app version and the newer ones can't be avoided at the moment, but we hope this is just a transitional problem. Occasional cross-platform validation failures might even happen for the same app version, but again, we are working on this, please bear with us.
Cheers
HB
Thank you Heinz-Bernd, it´s
)
Thank you Heinz-Bernd, it´s nice to see it gets attention and is being worked on, especially since I got another one last wednesday (WU 175297859). It´s a bit disconcerting having so many occasional invalids - as opposed to none before somewhen in August.
The last one I got invalidated against a Windows Server 2008 running BOINC 6.10.58 and a Linux box running a never official 7.0.36 IIRC; all v1.12 though. It shouldn´t matter of course, but it raises questions about the validity of the results, as it looks depending on engines: my now invalid result might have been perfectly valid if by chance, I had got the lottery prize of a wingman with the same host as mine? It is also a bit more unsettling because it happens on the bundled files taking about 15 hours here, instead of happening on the singles done in only 1h20 (assuming the mismatch would be on only one, not on all eleven).
Yet another invalid again, WU
)
Yet another invalid again, WU 174519025. Same settings as always now, v1.12 on all engines.
I could see it coming when my second wingman (the first timed out) raised ´inconclusive´ from his Windows host. The third wingman being Windows too, odds were not in my favor and so I have another 15 hours of crunching trashed. This is really becoming quite annoying, to say it mildly.
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/workunit/175651411