Consistent Early End On WU

[Tytalus]

I'm having an issue with my current SMP FAH (5.91) instances. For some odd reason the first instance (FAH #1 on my PC) will get to 49% (right on the edge of 50%) complete, and it'll EUE every time. This has been happening for a week now and I have no idea if I'm getting points for the early ends, but I'd much rather just have everything working properly.



Here's a quick screenshot--I didn't hunt down the issue but if anyone else is running 2xSMP on their quads they might want to double-check to make sure there aren't any issues!

Oh, and I have no OC at work, 4GB of RAM (restricted the SMPs to 1GB each, roughly) and the machine has been stable since I got it a month or so ago.

EDIT: Also, this was a Project: 2665 (Run 2, Clone 887, Gen 40) WU, in case anyone was curious.

 

[Rebel44]

Uninstall FAH

Delete FAH folder

Install FAH

 

[Tytalus]

Uninstall FAH

Delete FAH folder

Install FAH

Ouch! That sounds like a pretty wicked solution.... nothing simpler, like copying the DLLs and/or executables from my second folder over?

 

[Tigerbiten]

........Snip..........

Here's a quick screenshot--I didn't hunt down the issue but if anyone else is running 2xSMP on their quads they might want to double-check to make sure there aren't any issues!

Oh, and I have no OC at work, 4GB of RAM (restricted the SMPs to 1GB each, roughly) and the machine has been stable since I got it a month or so ago.

EDIT: Also, this was a Project: 2665 (Run 2, Clone 887, Gen 40) WU, in case anyone was curious.

Its easier to Copy-Paste the relavent parts of your FAHlog.txt file than to do a screenshot.
If its a long copy, just put it in code wrappers.

On to your problem.
Look in your FAHlog.txt file and check it was the same work-unit the errored out each time.
As it finished at 49% each time, you may well have a bad work-unit.
If the work-unit errors out, error b7, without sending any results back then you will download the same work-unit twice more before moving on to a new work-unit.
If you get really unlucky, you get the same bad work-unit for three more tries to fold it.

Look in your work folder.
Are there an orphan results files in there ??
If there are then use Qfix to try and repair the queue.dat file and make them returnable.
How to Qfix.
Once you've returned the results, then its easier to move on to a new work-unit.

If theres no ophan result files and you are still tring to fold the same bad work-unit.
Stop the client, delete your work folder and queue.dat file file, restart the client.
If you get the bad work-unit back, repeat the above step untill you get new work.

You can look up and report bad work-units Here

Ps. There are bad p2665's in the wild.

Luck ....................

 

[Pocatello]

Tigger Bitten is correct.

He says to use Qfix. He is right. It works. Very simple and quick. You can get all sorts of credit for doing so. It will report partial WU's.

Download it and run it.

Good luck!

 

[EvilAlchemist]

I would also suggest giving each client a little more ram.
Try 1.5GB and see what happens.

 

[Tytalus]

All pretty good suggestions, I'll give them a try later today or tomorrow! Thanks!

I probably won't bump up the RAM right now, since I have other considerations on this machine--I'll consider that a final option.

It just messed up another WU at 49/50%, which makes me think it's just a bum WU. I deleted the work directory and the queue.dat two days ago, so that didn't resolve my issue the first time. I could repeat it a few times and nuke some WUs until I get a different one, which I might try.

 

[Tigerbiten]

Before you nuke the work folder, your best checking in it for an orphan results file.
If you can send it back, Stanford gets info that this could be a bad work-unit.
The more feedback Stanford gets, the quicker the work-unit gets pulled.
Without any feedback, Stanford will keep sending it out, untill they get some.

Luck ..........

 

[Tytalus]

Before you nuke the work folder, your best checking in it for an orphan results file.
If you can send it back, Stanford gets info that this could be a bad work-unit.
The more feedback Stanford gets, the quicker the work-unit gets pulled.
Without any feedback, Stanford will keep sending it out, untill they get some.

Luck ..........

Turns out I sent back three units' worth of "partial completions" and got about 1500 points for them. Not too bad considering I was going to get 0 from them!

qfix was pretty sweet, and only took a few minutes to use.

The new unit I'm working on is now 51% complete (started ~24 hours ago) and things are looking better! It's still a 2665 WU but it's got different details, so I'm guessing it decided I shouldn't get any more of them.... which means no more torture for me!

EDIT: Bah! Can't lock my own thread, my GenMay sub is down for a month while I crush more proteins.

 

[relic]

I gave up on the SMP client. I am too busy to babysit the crappy programming.
I suggest you delete every 2665 you get if you catch it early. They're garbage.