Folding Forum

How many times does a work unit get put out to be done?

Obviously, only once isn't a good thing, unless the everythings fully trusted, but just how many people do the same wu?

AFAIK, everybody does a unique WU. the only exception is when there is an error with the client and that WU will be reassigned to another person. Also when the preferred deadline is over, the same WU will be reassigned to another person to complete. Other than these two cases, the WUs are unique to everybody.

In the most common scenario, an assigned WU is returned promptly and it will only be issued once.

Every WU must be complete, if possible, so if it is not returned before the Preferred Bonus expires, another copy is sent out -- assuming that the first one was lost. (Sometimes both are returned and both can get credit.) Although it's increasingly unlikely, this may happen several times.

Some WUs contain errors. Some hardware is incapable of completing a good WU successfully (commonly assumed to be due to unstable overclocking, but that's not the only possible cause). When a result is returned with an error, the server doesn't know if it's due to a problem with the hardware or with the WU itself. That WU is reissued, to see if someone else can complete it (i.e.- the first one was a hardware issue) or if it fails again (probably an inherent WU error).

WUs for FahCore_14 do not follow this pattern. Every WU is independent so even when the PRCG numbers are the same, the results are independent and useful.

Unlike other projects that must complete the same work unit more than once to assure data integrity, i.e. duplicate work and cut the project power in less than half, FAH uses other means to assure data integrity. As noted above, FAH does not duplicate work unless needed. It's part of the Run Clone Gen thing, among other things.

Well I have sure had the same one on 2 different machines at the same time and more than once...

road-runner wrote:Well I have sure had the same one on 2 different machines at the same time and more than once...

Yes, and with the already stated exceptions for errors, etc. Didn't you read the whole thread?

7im wrote:

road-runner wrote:Well I have sure had the same one on 2 different machines at the same time and more than once...

Yes, and with the already stated exceptions for errors, etc. Didn't you read the whole thread?

Those stated exceptions do not cover cases when one donor gets assigned a WU, and without the WU failing or exceeding it's deadline, he gets the same WU on different machine. (or different client.) While the clients were configured properly (different machine id), and they weren't FahCore_14 WUs.

7im wrote:

road-runner wrote:Well I have sure had the same one on 2 different machines at the same time and more than once...

Yes, and with the already stated exceptions for errors, etc. Didn't you read the whole thread?

Yes I read the whole freaking thread, I cant state my experiences? What the hell is wrong with you? You need to be laid or what, or are you still just trying to run people off? Man some peoples kids...

RR, stating your experiences doesn't seem to have ever been a problem for you. And thank you for bringing up the fact that Bruce's explanation doesn't explain EVERY duplication. However, we won't be able to explain EVERY dupe, as we are not PG. However...

^w^ing, one of the exceptions Bruce didn't mention is work server error. When a work server goes down, it's journal of WUs sent out may not be 100% accurate. So when the WS comes back up, it may resend a few of the same WUs as right before it went down to cover the bases.

Also note: Main FAQ: How do you set the deadlines for the work units?

Occasionally, deadlines may be set shorter... than the above [deadline] calculation indicate

This indicates to me that PG *might* at times set the preferred deadline rather short, causing WUs to be sent out more than once so that it gets returned very quickly. Whether accidental or on purpose, this can also cause a WU to be sent out more than once. If they need some WUs done really fast to meet a deadline, then send out 2 or 3 copies, and they'll likely hit at least one fast machine.

But again, as a general policy, WUs are not duplicated, or unless it serves a helpful purpose, or by error/accident. And if that doesn't cover ALL experiences, then sorry, we can't help explain them.

Yea no problem stating how I feel here, I dont know what caused it but it has happened, glad that it is not policy to do that though. No sense in running the same one through several machines wasting resources...

...Multiple instances

ihaque, Sat Feb 20, 2010 4:52 am wrote: Don't worry about it in the case of core 14 projects. The PRCG coding doesn't work quite the same way as for other projects; they're still doing useful work.

http://foldingforum.org/viewtopic.php?p=132228#p132228

^w^ing wrote:

7im wrote:

road-runner wrote:Well I have sure had the same one on 2 different machines at the same time and more than once...

Yes, and with the already stated exceptions for errors, etc. Didn't you read the whole thread?

Those stated exceptions do not cover cases when one donor gets assigned a WU, and without the WU failing or exceeding it's deadline, he gets the same WU on different machine. (or different client.) While the clients were configured properly (different machine id), and they weren't FahCore_14 WUs.

Really. Please explain how you know that somebody else didn't have an EUE causing the WU to be reissued? I've been trying to figure out how to tell from the data I can see and it's just not possible.

As 7im say, there probably are cases that we don't understand, but it's very difficult to prove that any such cases actually exist. It's important to differentiate between known established facts and speculation. (In this case, I include official documentation as established facts, even though I have not proven them myself.) I'm willing to speculate that the servers do make errors which depart from policy. It seems reasonable to assume that fixing a server bug which occasionally issues a duplicate WU is less important to the Pande Group than quite a few other bugs that might cause other types of server errors occasionally.

bruce wrote: Really. Please explain how you know that somebody else didn't have an EUE causing the WU to be reissued? I've been trying to figure out how to tell from the data I can see and it's just not possible.

I wouldn't know if somebody else had an EUE, but I can be pretty damn sure that I didn't have an EUE as I do have eyes, and I am literate so I can read my logs, thank you very much. So why would I get the same WU on a different machine?

Anyway I don't have the nerve for discussing things in the local manner. I've come to expect and tolerate the cold, un-involved, bitchslap posts from 7im (as he demonstrated himself very nicely in this thread by his response to road-runner), but the attitude seems to be spreading and I don't like that at all. Now, if you'll excuse me, I'll be on my way. Don't worry about me trying to add something to the discussions around here these days.

The point of confusion seems to be that, in the case of an EUE, the subject work unit is reissued multiple times. Therefore, you cannot tell from an instance of duplicate WU that something's wrong with the work servers. It could always have been someone else's failed unit.

I've noticed that when people ask the mods to see the outcome of an EUE, often multiple people show up in the records as completing it successfully. When someone inquires about a unit that went through, usually just one person shows up on the record.

Please do not confuse my brevity for anything else other than expediency. If road-runner would like an apology, I would be more than happy to provide one.

And unless I say that I am speculating or qualifying my statements with similar wording (like *might* may, possibly, indicates, etc.), my posted answers are not speculation. Even when I do speculate, it's based on empirical observations unless I down right say it's a SWAG. If the fact or the speculation in my post above are not abundantly clear, feel free to ask.

Wrish wrote:The point of confusion seems to be that, in the case of an EUE, the subject work unit is reissued multiple times. Therefore, you cannot tell from an instance of duplicate WU that something's wrong with the work servers. It could always have been someone else's failed unit.

I've noticed that when people ask the mods to see the outcome of an EUE, often multiple people show up in the records as completing it successfully. When someone inquires about a unit that went through, usually just one person shows up on the record.

Another good observation. We've all seen a WU fail 3 times before the client moves on to another work unit. If each of those EUEs generates another copy to be sent out, that easily explains some of the reported cases of multiple WUs, and the log examples Wrish noted.