• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Stanford Issues

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Sorin

Member
Joined
Mar 7, 2004
Location
Phoenix, since 03/2014
I was going to post this over at the folding community forums, but seeing as how I have the blank page login bug...

I've got a missing unit. It was turned in around 7:30pm PST, and still has not been credited, even at my stats page at Stanford. I read that while moderators can check for bad units, only the Pande group can check for missing ones. So since no one from there posts here or comes here, I think, I guess I'm just hoping the right person might see it. I know at least Mac and 7im (if he's the same one from FC) post here.

Anyways, it's a 2610, R/C/G 0/978/45. Results were successfully sent and the timestamp in the log for that shows 02:15:11. This was 7:30pm PST today, so I guess that translates to 2am on saturday, 8-18-07.

Anyone else got units missing right now? Or having that really annoying login bug?
 
I have no way knowing if I'm missing credits unless I miss a lot of them. I'll post in the forum for you.
 
From FC forums, by Bruce:

Sorin has returned the same WU twice. In most cases, this is caused by someone moving files from one computer to another without following the detailed recommendations for sneakernetting that can be found in the WIKI. In other cases it may be caused by cloning a copy of the Windows registry or by running more than one copy of FAH on a single computer without properly respecting the -local command line prompt or without properly setting MachineID.

See, I knew something was going to go wrong when I tried to swap units. I *did* read the wiki about "sneakernetting." But it only refers to moving units between two windows machines or windows-linux, and only when one computer has internet and another doesn't. The wiki doesn't say anything about just swapping work units between two machines that both have internet and are both Linux. I need to know exactly what has to be done in that scenario (two computers, both Linux, both with internet. Just a straight swap). I've searched on the internet multiple times in the past and asked more than once in the past, but I've never found the right answer. It's always just "move the work folder and queue.dat," when judging by the wiki info, it's clearly more complicated than that.

What I was referring to in my initial post in the other thread I started about project/assignment server/machineid mix-up is here.

At this point I'm thinking it may have something to do with machinedependent.dat, which I think is the Linux equivalent to the machineid info in the registry for Windows.
 
If you sneakernet a WU, the originating machine is going to go back to the AS without having returned the last WU assigned. The AS is programed to give you the same WU again unless you are past x% of the prefered deadline (I have no idea what x is and I'm not really sure what the logic is, but % PD of the original WU makes sense). The work around is to set up two or more instances on the originating machine so that after transfering the WU, you use a different instance for the next WU. You'll typically see large numbers of active cpus in the last 50 days compared to 7 day actives on users doing a lot of sneakernetting. Alternatively you can watch the console and delete the work dir and queue.dat if you're assigned the same WU you transfered and keep doing it until you get a new WU. On the 5.0x clients you can usually add or delete -advmethods on the originating machine to avoid assignment of the same WU.

I don't think you'd have this problem if you swapped WUs between two machines, but I haven't tried it.
 
I have swapped a few linux wu's and it has mostly been ok.

As chasR pointed out, i try to double check each involved instance that it's next wu is not a duplicate. Most of my sneakernet scenarios with smp have been on linux quads where on some I often run 2 instances already ... so when i get 2 incompatible wu's on the same rig (2608/09/10) then i will sometimes swap one to another linux rig. The alternative in my scenario is to run the 2 instances serially ... ie stop the least complete one to let the other finish with a -oneunit or -pause on the launch string and then run the other one.
 
Last edited:
The only occasions I've had recently for sneakernetting a WU has been when a machine is permanently going out of service due to hardware upgrade, or when the o/s has been replaced( ubuntu to debian).
 
ChasR said:
I don't think you'd have this problem if you swapped WUs between two machines, but I haven't tried it.

That's exactly what I did though. Just swapping between two machines due to unfriendly work units on the same box, as Pete mentions --->

pscout said:
I have swapped a few linux wu's and it has mostly been ok.

As chasR pointed out, i try to double check each involved instance that it's next wu is not a duplicate. Most of my sneakernet scenarios with smp have been on linux quads where on some I often run 2 instances already ... so when i get 2 incompatible wu's on the same rig (2608/09/10) then i will sometimes swap one to another linux rig. The alternative in my scenario is to run the 2 instances serially ... ie stop the least complete one to let the other finish with a -oneunit or -pause on the launch string and then run the other one.
 
I guess the AS doesn't care if you return a WU not assigned to a machine, assumes you've lost the one it assigned, and reassignes the same WU you just swapped away. If you happen to swap both soon after download, both machines will likely download the WUs swapped away. Two instances on both machines along with the -oneunit flag will fix that. Or sitting there and watching what gets assigned and deleting duplicates untill you get a new WU.
 
Back