Unexpected Shutdown problems - will they cause the new client major problems?

[tpdooley]

For close to 6 months I've had 1-12 machines with 24/7 internet connections (on dialup) folding away. I've had some of them lose power and shut down unexpectedly, I've turned at least one off, I've rebooted one while on the wrong keyboard, and I've had a few windows updates end the update by instantly rebooting the system.
In all those cases, I lose the information on the 10k structures being worked on. Booting back up, I restart DF, and it continues on (while I grumble about losing 200-9800 folds.. and those higher values do hurt a bit..

When folks are running /nonet (by choice or not i.e. the cable modem needed to be rebooted to pick up the latest change in dns info) - and the system is unexpectedly shut down, they get messages similar to:
FATAL ERROR: [000.000] {foldtrajlite.c, line 760} Illegal file found in upload list
and we have to edit filelist to get rid of the entries that DF is hanging on - to get most of the stored collection to upload.

can we get an official program aka "crashedtest.exe" - that would rebuild the filelist, insuring that the matching set of .log.bz2 and .val.bz2 exist, and that they're in valid form and hold valid data?

----------
For the new algorithm - where our current results are based on the results of the previous generation - can you make the client a little more resilient to unexpected shutdown? to store temporary collections of 100 folds (with the last 100, plus the preceeding 4 collections) - eliminating the oldest with a successful write of the newest) - so that at any crash on generation 49 (where things will move relatively slowly) - that we don't lose significant work?
(please balance the resiliency with speed of production - since if the resiliency features slow the client down by 1k structures a day, most of us would be better off running full bore, and just grumbling about the structures lost - since this is a fairly infrequent situation).

[Brian the Fist]

These are the sorts of issues that we hope to address in the beta testing of the new algorithm. If you wish to have a say, please participate as a beta tester (we will advertize here in the forum, shortly, when we are ready to beta test).

[tpdooley]

Guess it's time to setup another system so I can test out the new algorithm, then.

*grin*

[runestar]

I do believe that is a long-running concern of many people. Although at this time we are just sampling known proteins, and the loss isn't critical, at some point when DF gets into the actual definitive predicting of folding patterns, it will become critical not to unecessarily lose them. Of course there will still be the mishaps, sometimes a wierd one which will cause a loss.


BTW, I'd be happy to beta test again Howard both scrsaver and cmd-line clients. Maybe you should release the screensaver first so they have to use that first. Your team went to all that hard work making it and no is even taking a peek at it.

Maybe you can sign a deal with AOL to distribute it with their CDs. =) You just have to give them out on campus... but no will mind... we get them in our cereal anyways. Hey there's an idea, Scott. If Google can do this whole distributed computing power thing deal with Folding@home, why shouldn't Howard do it with AOL? They've got the largest base of subscribers of anybody. If we could get just 500k AOL users using the screensaver... That would shake the big teams up... an AOL team. No self-respecting geek team is going to let crappy AOL upstage them... =)


TTFN,

RuneStar½