[QUOTE=TheMawn;366565]How hard would it be to give mfactx a more friendly interface like what Prime95 has? Or maybe include GPU's into that interface?

I should stress I don't want to sound like I'm demanding more work being piled on. Seriously just asking.[/QUOTE]

You could try if you wish :smile:

Ok, on more of a serious note, people here are really like: "If it works don't fix it"

[QUOTE=chalsall;366532]Please don't, however, implement the new "recycling" rules for another couple of weeks, as this will result in a bit of a surge for Cat 4 requests that I'm not sure we're ready for.[/QUOTE]

Hey George et al.

We now have over 1,600 candidates above 66.2M TFed to at least 74 "bits" and P-1'ed (and another 400 already out being "worked"), so I think it's safe for you to implement the new assignment and recycling rules whenever you'd like.

[QUOTE=TheMawn;366565]How hard would it be to give mfactx a more friendly interface like what Prime95 has? Or maybe include GPU's into that interface?

I should stress I don't want to sound like I'm demanding more work being piled on. Seriously just asking.[/QUOTE]
GMTA?

Last night I started some design work on a utility like this. Except I'm not doing it by modifying mfactx, but rather having it use mfactx as a plugin. In theory it would mean that it could provide a more-friendly and automated interface to any tool, so long as you have the plugin to link your given testing program to it.

But enough hijacking this thread....

[QUOTE=chalsall;366755]

We now have over 1,600 candidates above 66.2M TFed to at least 74 "bits" and P-1'ed (and another 400 already out being "worked"), so I think it's safe for you to implement the new assignment and recycling rules whenever you'd like.[/QUOTE]

New LL assignment rules are now in place.

Note on the thresholds page changes to requirements for both DC and LL assignments. It will be hard for a computer running all day to not qualify for a category 3 assignment.

[QUOTE=Prime95;366824]New LL assignment rules are now in place.[/QUOTE]

Thanks George! Monitoring the load.

Next, of course, is enforcing the recycling rules. IMHO, [URL="http://www.mersenne.org/assignments/?exp_lo=45000000&exp_hi=50000000&execm=1&exdchk=1&extf=1&B1=Get+Assignments"]this report[/URL] shows clearly why these new policies Makes Sense [SUP](TM)[/SUP].

(And, look who we find holding four of the lowest candidates within the bottom 15... Less than 40% complete after more than a year.... :sad:)

Agreed. Kind of sad to see that 5 million exponents worth of milestones are held up by a few dozen stuck assignments. Most of the ones grabbed in late 2013 or early 2014 are on par to be finished quite soon. I'm too busy to look at that in too much detail but it seems like most of the recent assignments would meet the criteria we've established.

[QUOTE=chalsall;366854]
(And, look who we find holding four of the lowest candidates within the bottom 15... Less than 40% complete after more than a year.... :sad:)[/QUOTE]

Maybe this is a "thing which makes one go "hmm".

I'm almost in favour of the BitTorrent approach: when the last block is slow, also request it from a fast peer and get it done. Heh.

Houston, we have a problem...
 

George et al...

While the new DC assignment rules transition appears to have gone quite well, the LL transition, um, not so well...

[B][I][U]Very[/U][/I][/B] few candidates in Categories 1, 2 nor 3 are being issued; almost all new assignments are in the Category 4 range, and very quickly our supply of "TF'ed to 74, P-1'ed" candidates were exhausted in that category.

I suspect this is because few of the "Big Iron" LL'ers "Promised to complete assignments quickly" (including, for example, curtisc and Amazon EC2). I would suggest you (George) disable the new assignment rules until we work out a solution.

One idea would be to set the "Promise" field for the top 100 LL producers (except for non-known (rather than simply "hidden") "Anonymous" accounts), and also possibly set the "Number of Days to Queue" value to be 10 for historically fast machines, and 30 for medium fast.

Also, for anyone here who does LL work and hasn't already, you should make your explicit "Promise" at [URL="http://www.mersenne.org/thresholds/"]the Thresholds page[/URL].

I've disabled category 4 until I can figure out why there were so few category 3 assignments. I had assumed big players like curtisc would get mostly category 3 assignments.

Hmm, in looking at assignments above 67M since Feb 14 it looks like everything was operating as expected.

There were 2062 assignments of which only 88 went to computers that had previously submitted a result.


So what might be happening here and is it a bad thing? Are there a large number of stress testers inadvertently signing up for tests that will expire 60 days from now? This wouldn't be happening in the DC area.

If this theory is correct, we are essentially building a buffer (pre-reserving) exponents below 66M for the reliable machines to finish over the next year. For the next 60 days, there will be a surge in reservations above 66M few of which will actually be completed.

One idea would be to phase in the 100000 limit. Say 60K this month, 80K next month, 100K the month after.

Does anyone have any other theories or suggestions?