[QUOTE=Prime95;432325]See post #2415[/QUOTE]

So were both cat 0 M67122481 and cat 1 M67542281 assigned in error? In any case, the cat 1 (M67542281) should get recycled around May 6 or May 7, assuming the owner doesn't report starting it.

[QUOTE=Prime95;432325]See post #2415[/QUOTE]

Being no direct answer to Jacob's question was provided, I'll just say that post #2415 has two parts. If it's the first part that we're supposed to "see," the implication would seem to be that the machine in question has been productive recently, but for some unknown reason, a ridiculous ETA was calculated for a new exponent. If it's the 2nd part of #2415 that we're supposed to "see," then the implication would seem to be that [url=http://www.mersenne.org/M67542281]67542281[/url] was assigned in error (in my view, a more logical explanation).

The latest stats for 63350927 (my own analysis) shows that it will finish in 20.5 days, just shy of the 18-19 days before it expires.

We're faced with one of those dilemmas again where it could be a real squeaker. The pace has picked up a bit ever since it stopped running at all for about a week, so it may actually recover enough oomph to finish in time.

On the other hand, if it might only miss the expiration deadline by a day... should we intervene and manually extend it? I guess that's a good question for George... personally I'd probably extend that courtesy since it really will be a difference of 24-48 hours.

There will be 18 more days for it to make some improvements in the predicted ETA (the client itself is still reporting an ETA that's 4 days further out than my own prediction, but Prime95 is pretty conservative in it's estimates). I just wanted to throw out the idea that this might be a good candidate for extending it a bit and letting the original assignee finish.

The alternative to letting it expire is that someone else will get it, start work on it, and unless they have a super fast machine, they'll be "poached" a day or two later by this expired work. Not as big a deal for first time checks, true, but still annoying for the person who just got the new assignment.

We could probably do with something in the expire rules. If an exponent is x% complete and will complete within x days then it is extended by x days as long as it reports in every x days and continues making progress.

It seems to me that a fair few exponents are coming up that expire but will probably be finished before the next assignee finishes it. We should be looking at the progress made.
To me if an exponent which has 60 days to complete is 70% complete at 60 days then it shouldn't be expired as the remaining 30% will probably take less than 30 days(quicker than we would as the next assignee to do).

[QUOTE=Madpoo;432584]The alternative to letting it expire is that someone else will get it, start work on it, and unless they have a super fast machine, they'll be "poached" a day or two later by this expired work. Not as big a deal for first time checks, true, but still annoying for the person who just got the new assignment.[/QUOTE]

I would argue you let "nature" take its course. Keep in mind it actually has 19 days to complete, and has done 3.1% in the last two days. And if you make an exception in this case, do you also do so for his other seven assignments? What about others in a similar situation?

The risk of a Cat 0 being "poached" by the expired assignee must be a risk accepted by those explicitly opting for that work type.

Fortunately soon this type of issue will occur far less frequently, as users like Summy won't be given inappropriate work even though they've "promised".

[QUOTE=chalsall;432589]I would argue you let "nature" take its course. Keep in mind it actually has 19 days to complete, and has done 3.1% in the last two days. And if you make an exception in this case, do you also do so for his other seven assignments? What about others in a similar situation?
[/QUOTE]

I would agree. And the advantage would be that that exponent gets doubled checked in a very short time. (and in the hypothetical that the both residues do not match we are made aware of this sooner rather than later)

Yesterday the last unassigned DCs under 36M were assigned. ~400 to go.

[QUOTE=henryzz;432588]We could probably do with something in the expire rules. If an exponent is x% complete and will complete within x days then it is extended by x days as long as it reports in every x days and continues making progress.[/QUOTE]

We should consider this.

First of all, I'd give the 2-day extension if it looks like it will complete. It isn't that I worry about hurting his feelings, rather I don't think it is fair for the re-assignee who has signed up for first-time tests.

In this case it is no big deal as no work is really wasted, but for DCs it becomes much more important. Right now the database only keeps track of the last progress report and thus cannot make an accurate prediction of the completion date. Madpoo either has a SQL table I don't know about or is tracking it manually.

Maybe madpoo can comment on the feasibility of making accurate predictions and then we can discuss suggested rules for grace periods.

One idea is to require a registered e-mail for users who want Category 0 or 1 assignments. This will allow PrimeNet to send reminders when assignments are close to expiration.

[QUOTE=Prime95;432610]It isn't that I worry about hurting his feelings, rather I don't think it is fair for the re-assignee who has signed up for first-time tests.[/QUOTE]

Agreed... that ^^^

I'm more worried about the new assignee getting "poached" (but not really) when this finishes up only 2 days into their own run.

[QUOTE=Prime95;432610]In this case it is no big deal as no work is really wasted, but for DCs it becomes much more important. Right now the database only keeps track of the last progress report and thus cannot make an accurate prediction of the completion date. Madpoo either has a SQL table I don't know about or is tracking it manually.

Maybe madpoo can comment on the feasibility of making accurate predictions and then we can discuss suggested rules for grace periods.[/QUOTE]

Indeed... a SQL table I setup to track a daily snapshot of all assignments that have updated since the last time and have some progress to report. If they've checked in but the needle hasn't moved since last time, I think I ignore those.

Anyway, then it's down to an analysis of a particular exponent's progress over time which lets me estimate the actual daily rate.

It can be thrown off by periods of inactivity (machines being turned off for a day or two), but it tends to average out over long runs.

Once an assignment checks in, the history for it is removed as a way of cleaning up the DB. Expired assignments might get removed too...

For M63350927 specifically, the latest report shows it expiring in 15 days and finishing in 16.9.

Assignments get a sort-of reprieve since the expiration rules run just before midnight, so realistically it would expire about a day before it finishes (if I'm remembering that right).

Seeing that the demise of the 36M milestone draws nigh, perchance Aaron would find it expedient to forthwith codify the placement of 37M to list of future achievements.