PDA

View Full Version : Caching WU's

[XC] riptide
05-12-2007, 04:13 PM
Is there a way to cache WU's for Windows GPU client. I have not always on internet and was wondering can I Q the WU's both finished and 'ready to run'.

Had a look around here and didn't see much.
DAK1640
05-12-2007, 04:27 PM
I too would be interested in this caching...
[XC] riptide
05-12-2007, 06:15 PM
http://www.distributedcomputing.info/tools.htmlMaybe something in here.... UD monitor?
SparkyJJO
05-12-2007, 06:52 PM
I'm really not sure. I've heard that there is a way to do it but I haven't figured it out nor really looked into it much.
jimwah
05-13-2007, 02:58 AM
6 day deadline means you'd only be able to cache a few units really, Stanford want a quick turn around on these GPU units - as with the HT issue, the quicker the results get back, the happier Stanford are apparently.
mike047
05-13-2007, 03:41 AM
At one time FAH had the "tinkers" timeless wu's that could be queued 10 at a time.[and sneakernetted:D] They discontinued their use and I left FAH almost completely[2boxes still]. It made for some real trouble free crunching, oh well.
Scimitar
05-13-2007, 10:12 AM
Incorrect. Tinker work units are still being used. They are just a small minority compared to Gromacs units now. Take a look at the current projects list on Stanford's website for details. I folded one recently.
mike047
05-13-2007, 12:10 PM
Incorrect. Tinker work units are still being used. They are just a small minority compared to Gromacs units now. Take a look at the current projects list on Stanford's website for details. I folded one recently.

As usual, the context of my post was not understood. Sorry if anyone thought that "tinkers" did not exist anymore...obviously they do as pointed out.

At one "time" they were in abundance:D and one could crunch them exclusively and sneakernet then very easily. They were queueable in 10 wu groups.

Is the tinker you did available in 10 unit groups:)

MY FAH (http://vspx27.stanford.edu/cgi-bin/main.py?qtype=userpage&username=mike047)
[XC] riptide
05-13-2007, 03:20 PM
Mike just scored a PWN!
[XC] riptide
05-17-2007, 03:30 PM
UD monitor seems to be working for me.
[XC] riptide
05-29-2007, 02:53 PM
Right! After a week messing around with it I would happily recommend UDmonitor as a Folding GPU caching and client control app. It will handle CPU folding instances, but it only seems to be any good for single instances.. or single core machine.

It switches nicely from one WU to another as they finish. And you can flush and renew the cache at your leisure.

One thing to note, if you get an early unit end from a crash/high overclock it will stall at the 'trying to contact server' stage and that Folding instance will stay active until it goes through its normal upload and download.. ie: it won't switch over to another WU. And that is prolly a good thing. You need to get back to the clicent and sort out the HW before you mess other WU's up.


Again.. this is all with a not-always-on internet connection. And for 24/7 GPU folding it works well.