Newer milestone thread

[TheMawn]

Originally Posted by Brian-E

Is there any particular reason why long-standing no-progress work in "preferred areas" is not reassigned? If it's a time and resources issue, can any particular trusted person perhaps be officially nominated to manually redistribute this work under very clear project guidelines?

I completely agree. And I'm sure half the people in this discussion have thought up of their own quite functional guidelines.

My own thoughts: Let the smallest 5000 exponents of the LL and of the DC range be called the trailing range for each assignment type. Trailing-edge exponents are only assigned to registered users whose rolling average is high enough that it would take them less than two weeks to finish for a DC, six weeks for LL.

If reported progress is such that it would take three or nine weeks, respectively, to finish, the assignment can be recycled at any time.

This is just so that the bottom can be cleaned up a little bit. The actual numbers are purely for example. Maybe they're a bit optimistic. I don't know.

Originally Posted by tha

I believe quit some of these troubled assignments are done by PC's that have been given a different role by it's owner later. Like primary PC replaced by new one or a tablet in the household that makes the user change it's always on habit.

This kind of goes hand in hand with the idea that a bunch of people who set up Prime95 to start on bootup probably forgot that Prime95 runs at all, and are actually just being a hindrance by running an assignment just long enough to refresh the time limit.

Honestly, there's all this talk of changing how the assignment ownership setup works with "preferred" assignments but I think just killing any assignment with no work done in six months or progress so slow it would take years, would clean things up a lot. I think that should be the first step.

[garo]

The last big push to clear out assigned but not worked on exponents came about 10 years ago. A forum member who went by the name of GP2 compiled a list of exponents that were unlikely to be finished any time in the near future and had been assigned for 1.5 years or more. After much consultation with forum members - and carving out legitimate exceptions from the list - George beat the v4 server till it coughed up these assignments and everyone lived happily ever after.

<movie trailer voice>Until now</movie trailer voice>.

I propose that someone make a list of exponents that have been assigned for > 720 days for LL and > 360 days for DC, then prune exponents that look like they are going to finish in the next month or so based on their recent progress. Hand this list to George, sit back and watch the magic happen.

[garo]

Here is the thread y'all need to read to see the kind of data that prods George into taking action . http://www.mersenneforum.org/showthr...&threadid=1201

And the follow-up here:
http://www.mersenneforum.org/showthread.php?t=1381

GP2's research on error prone machines led to significant improvements in the way exponents with non-zero error codes were handled. v5 has the LLERR column thanks to this work and releases 1st time exponents with non-zero error codes for re-testing immediately.

Look at http://www.mersenneforum.org/forumdisplay.php?f=35 for the last time forum members tried to clear out blockers in a coordinated fashion.

The thousand day club:
http://www.mersenneforum.org/showthread.php?t=1349

GP2 we miss ya!

[TheMawn]

How did these mystical men of the past find all that data? It seems like there were four or five really really big .txt's floating around or something.

Originally Posted by JC

If I remember well, first 33M exponents have been released in September 1999. At that time, one full year was needed by top-speed machines to complete them.

LOL

[Brian-E]

Originally Posted by TheMawn

How did these mystical men of the past find all that data? It seems like there were four or five really really big .txt's floating around or something.

Yes, there were. For example, you could download a text file with all current LL test assignments and how far each had progressed - to the nearest interation! - at last report. By collecting a series of these files over a period of time you could quite easily write a quick program to detect old assignments which were showing no progress. I believe those text files stopped being available with the migration from v4 to v5 in 2008. Now I guess we have to construct suitable database queries.

[chalsall]

Originally Posted by Brian-E

Now I guess we have to construct suitable database queries.

Yes. It's become a little more complicated since v4 of the server.

"Spidy" is already collecting the required data. I'm working on a Perl script which will answer at the least the "real-time" question of "Assigned Date" - "Claimed Estimated Complete Date".

[kladner]

Settin' on top of the world

[petrw1]

Originally Posted by TheMawn

How did these mystical men of the past find all that data? It seems like there were four or five really really big .txt's floating around or something.

LOL

In V4 there was a readily available file listing ALL assignments.
Granted that was only up to 79.3M then.
and P-1 and ECM were NOT listed types.

Am I showing my age?

[TheMawn]

Originally Posted by chalsall

Yes. It's become a little more complicated since v4 of the server.

"Spidy" is already collecting the required data. I'm working on a Perl script which will answer at the least the "real-time" question of "Assigned Date" - "Claimed Estimated Complete Date".

Cool. I would have been first in line to help but my programming experience is one Visual Studio C++ course I took two years ago. I'm not too bad with Excel, though, if you need a hand processing the data once it's been gathered.


Regarding the team 1000+ entries, I think a lot of those were semi-decent computers working on exponents larger than the ones by a factor of six or seven. I don't think we ought to care how long someone's been working on some 100M exponent until the trailing edge of the wave is there.

[chalsall]

Originally Posted by chalsall

"Spidy" is already collecting the required data. I'm working on a Perl script which will answer at the least the "real-time" question of "Assigned Date" - "Claimed Estimated Complete Date".

OK, my first attempt at filtering this data is now complete. A CSV format file is available at https://www.gpu72.com/data/20140119_...ree_months.csv

It contains all the current assignments (as of today) which are 90 days of age or older. This is intended to be imported into a spreadsheet program (or other custom code) for sorting and drill-down.

Hopefully the column headers make sense; the "Need" column is "0" if the work being done has already been completed.

The "Est_Comp" column is the number of days after assignment the candidate is expected to take to complete, based on the percentage actually completed combined with the age of the assignment. If no percentage completed has been reported (common for Manual Assignments) then the value is 999999999.

Note that candidates under the control of GPU72 for TFing and/or P-1'ing is not included in the list, since we marshal their progress very carefully.

Please let me know if anyone has any comments, questions or suggestions.

[kracker]

Well well, look who we have on top

EDIT: Hmm.. I see I'm in there. I didn't know I had LL assignments outside gpu72... Wonder why