Folding Forum

When I want to check my stats on kakao I found a folder (Angra) who made in the last 24 hours 788,864 points. Then I checked the official Stanford stats site because I want to know how many clients he is running. I was very surprised when it shows only two active clients for the last seven days.

What type of supercomputer do this guy run?

Something is fishy here. AFAIK, the maximum PPD is obtained by bigadv WUs so a 48-Core system gives 171,631 PPD (Details) and if there are two instances of it, it nets 343,262 PPD which is still less than 788,864 Points

This one is very suspect ... or something is going very wrong with the stats

He has two active clients (1 NV GPU and a BigAdv one), but I see multiple submissions of the same WU from his Linux machine ... I don't know how it's possible, but he got credits 10 times for the same WU (with very similar characteristics in the result report).

I'll ask for more investigations from the Pande Group ...

Before labeling a cheater you might want to check with PG first. Perhaps he was just back-credited for a bunch of WU's that didn't get proper credit earlier. It happens...

I edited, do you like the wording better ?

P.S : if it was late credit, I guess he wouldn't have submitted 10 times the same WU ...

http://www.xtremesystems.org/forums/sho ... stcount=12

http://www.xtremesystems.org/forums/sho ... ostcount=7

The above shows more than two clients folding. With the duplicate WUs it appears the config has been cloned from one client to another or something to do with clustering but that doesn't explain why they would be getting full credit for duplicates

OK. 18 running clients would explain his ppd, even though this screens seems to be from September. And his machines looks well equipped (~ 7h for a 6701 WU).
I'm curious about what PG will say.

If you read through some of his post he is running some kind of a cluster and he has recently started up 18 Duel X6 servers in that cluster. I do not know anything about clusters but I am assuming that could account for only 2 folding machines showing up on the official stats. But I do not understand the 10 duplicate WU thing.

http://www.xtremesystems.org/forums/sho ... p?t=256990
http://www.xtremesystems.org/forums/sho ... p?t=258551

We'll take a look.

Configuring 18 systems with the same User id/ Machine id would look like only one or two machines and would produce a lot of duplicate WUs. The real question is why he got credit for so many of them. He should be getting the message about the server already having received the WU and the other guy who keeps getting the message should not be getting it.

Indeed ©

I just happened upon this thread while looking for something else. I don't read over here often.

I was not aware that having several computers with the same machineid would produce duplicate WUs. I was always under the previous impression that I only needed unique machineids for multiple clients on the same machine.

In fact I did start getting duplicate WU errors several months after this thread, which prompted me to reinstall the clients on all my machines. Since no one has complained of fishiness since then, I'm guessing that fixed the issue.

Is there any way for me to see how many duplicate WUs I ended up getting credit for? I was not aware that there was a problem - I'm surprised no one from here posted in any of the threads linked above.

Different machines can have clients with the same machine id (1,2,3 etc), but the clients on one machine must have different machine ids.
From F@H FAQ - Is there anything special I should do to run on a cluster?

The main aspect to consider is to make sure that the CPUID for each machine is unique. To help fight from having duplicate IDs, the Windows versions (v3 and later) keep their IDs in the registry and the Linux version (v3.11 and later) keep it in a special file MachineDependent.dat.

Ways to avoid duplicating IDs:

1 If you install each client individually, then it's impossible for there to be problem with a duplicate ID.
2 If you use the recent Windows version and have single processor machines, then you should also be fine (for duals, see the above).
3 For a Linux cluster, make sure that if you copy the directory, you DO NOT copy the MachineDependent.dat file. This file will be auto generated by the client to get a new ID.

A Mod will have to check how many duplicate WUs you've been credited for, but it sounds like you're not doing that anymore.

Right - OK, that all makes sense. I don't think I've read the FAQ in 5+ years, since I initially set up my farm.

What I'd really like to know is how much worthless work did I do, and if some thought I was cheating, or even just misconfigured, why I kept accumulating points without anyone complaining at me. I do this project because I like what F@H is doing - I don't want my cycles doing duplicated work

Instead of putting chunks of information in some FAQ and blaming it on the user if something goes wrong you also could have gone all the way with the client and have checks built in to make it really unreasonable for more than one machine of one user to have the same CPUID. Check MAC of eth0 or whatever and have the client save that bit of information. If you copy over the whole client directory to another machine, which is presumably on the same network, MAC differs and client generates new CPUID. Problem solved.