Newer milestone thread

[chalsall]

Originally Posted by Madpoo

I'm not sure when to expect that one to be in the critical range based on the current progress, but it has some catching up to do.

Approximately 130 days from now based on our ~67.7 Cat 1 completions a day average over the last 30 days.

[Madpoo]

Originally Posted by ATH

This is what I suggested a while back except I suggested the average over last 3 checkin to better average out peoples irregular behaviour:
http://www.mersenneforum.org/showpos...postcount=1548

I actually meant the 4th last checkin in that formula so we get 3 gaps between 4 checkins, but 3 is just a suggestion maybe another number of gaps is better.

My first pass at this is as follows:

• Once a night (just past midnight, UTC) I take a snapshot of the progress of all current assignments, if they've checked in during the past 24 hours. I may adjust that to exclude work with 0% complete since it ends up being kind of meaningless.
• That info goes into a new database I setup that clears entries older than 30 days or when the assignment is removed.

Currently, I don't have a ton of data to work with so far... to work at all I need at least 2 check-ins so I have a reference point and then a point of comparison. As I noted, some of the entries have 0% as the work complete, and that may be from an assignment that hasn't actually started yet or was just assigned. Including work that hasn't yet begun is only going to throw off the progress so I'll probably remove those entries and keep them from being added in the future.

I've done some basic analysis of a few interesting assignments. For example, 55771997 and 55861261 since those are in that group of first-time checks below 56M awaiting completion.

Even though I may be capturing multiple data points along the way, if they check in several times over 30 days, my initial go at this is just taking the first and last dates and comparing the % complete recorded.

In those 2 cases, we're only talking about 48 hour intervals so it may not be totally accurate yet, but it should get better as more data is recorded.

Anyway, to the nitty gritty:
Each of those has recorded a rate of progress of 0.05% every 24 hours. 55771997 still has 13.8% to go so it's "real ETA" is more like Jan 15, 2016. 55861261 is moving at the same rate but since it only has 1.3% to go it should complete on May 10, 2015.

Since my previous look at things shows that even grandfathered first time checks will expire regardless when it's over 665 days old, unless that machine picks up the pace, it will expire before completion. Someone with more time than me could probably see where the lines intersect and pick the exact day that would happen, assuming it continues at a linear rate.

At least the second one will probably finish up soon.

As I mentioned, this is preliminary since those systems have only checked in twice since I started gathering data, and the 0.1% progress it made in 2 days between check-ins may go up or down as it averages out.

As I look at some other assignments out there, there are some "outstanding" ones, and I mean that in a bad way. :)

Worst case is 36292681 which has progressed from 0.2% to 0.3% in about 34 and 1/2 days. At that rate I'm projecting it'll finish in the year 2109.

Eventually I hope I can do a follow up analysis and see if my predicted ETAs match the actual time it was finished, but since a client may complete it's assignment and then check in it's results much later (either manually or because it's not connected to the 'net 24/7), I don't know if I'd worry about that too much.

I already see a few cases where my amazing prediction machine is saying some assignments *should* have been finished yesterday but they're still not done, so I'm not sure if that's just because the client hasn't bothered sending it in yet or because my data set isn't really that accurate yet, etc.

[Madpoo]

I'm looking specifically at the first time tests between 54M and 56M (just those last 8) that we're waiting for.

One of them hasn't checked in again since I started gathering stats, so I have nothing to work on with it.

The other 7 in that range have checked in at least twice which gave me enough of a running "throughput" to make the following best guesses:

Exponent   "Real" ETA
--------   ----------
54674791   2016-03-17 20:18:11.823
54759797   2016-03-26 00:59:11.540
55027163   2015-10-26 11:08:38.387
55059383   2015-10-05 07:10:38.590
55079077   2015-10-28 16:12:38.790
55771997   2016-10-19 13:24:44.550
55861261   2015-06-07 11:08:45.013

That's all based on their percentage to go and their observed "percent per day" rate.

[Gordon]

Originally Posted by Madpoo

I'm looking specifically at the first time tests between 54M and 56M (just those last 8) that we're waiting for.

One of them hasn't checked in again since I started gathering stats, so I have nothing to work on with it.

The other 7 in that range have checked in at least twice which gave me enough of a running "throughput" to make the following best guesses:

Exponent   "Real" ETA
--------   ----------
54674791   2016-03-17 20:18:11.823
54759797   2016-03-26 00:59:11.540
55027163   2015-10-26 11:08:38.387
55059383   2015-10-05 07:10:38.590
55079077   2015-10-28 16:12:38.790
55771997   2016-10-19 13:24:44.550
55861261   2015-06-07 11:08:45.013

That's all based on their percentage to go and their observed "percent per day" rate.

October 16?

Doubt that will survive, what chance somebody "accidentally" types that exponent into a manual check....

[NBtarheel_33]

Exponent   "Real" ETA                 665 days old on
--------   ----------                 ---------------
54674791   2016-03-17 20:18:11.823    2015-09-11
54759797   2016-03-26 00:59:11.540    2015-09-11
55027163   2015-10-26 11:08:38.387    2015-09-05
55059383   2015-10-05 07:10:38.590    2015-09-05
55079077   2015-10-28 16:12:38.790    2015-09-05
55771997   2016-10-19 13:24:44.550    2015-10-12
55861261   2015-06-07 11:08:45.013    2015-10-12

[Madpoo]

Originally Posted by Gordon

October 16?

Doubt that will survive, what chance somebody "accidentally" types that exponent into a manual check....

Well, I know there are some out there (and I guess I'm one of them) who don't mind poaching an assignment that seems abandoned (and yes, I've been wrong before because I wasn't paying attention).

It wouldn't be much of a stretch to say poaching one that is just going to expire anyway before completion, even though it's being slowly worked on (as little as 0.1% in 2 weeks in some cases), is probably okay too.

And now I should probably duck and hide since not all would agree. LOL

[ATH]

I have poached many exponents in my time so I'm not one to talk, but maybe lets try and see if the recycling system actually works and not poach them just before?

As I wrote 2-3 times in this thread the 2x 54M exponents will "expire" in mid/end of May 2015, so only ~ 1 month to go.

[Gordon]

Originally Posted by ATH

I have poached many exponents in my time so I'm not one to talk, but maybe lets try and see if the recycling system actually works and not poach them just before?

As I wrote 2-3 times in this thread the 2x 54M exponents will "expire" in mid/end of May 2015, so only ~ 1 month to go.

I was referring to the October 2016 ones...

[ATH]

Originally Posted by Gordon

I was referring to the October 2016 ones...

The October 2016 dates was the ETA on them finishing if left alone, but they will not be left alone they will be recycled way before that, the 5 lowest of 7 exponents should be recycled within 2 months:

Exponent   "Real" ETA	665 days old	Recycled on
--------   ----------	------------	-----------
54674791   2016-03-17	2015-09-11	2015-05-25 + 3.33 days for every % above 67.00%
54759797   2016-03-26	2015-09-11	2015-05-22 + 3.33 days for every % above 66.20%
55027163   2015-10-26	2015-09-05	2015-06-11 + 3.33 days for every % above 73.90%
55059383   2015-10-05	2015-09-05	2015-06-08 + 3.33 days for every % above 72.90%
55079077   2015-10-28	2015-09-05	2015-06-10 + 3.33 days for every % above 73.60%
55771997   2016-10-19	2015-10-12	2015-08-27 + 3.33 days for every % above 86.20%
55861261   2015-06-07	2015-10-12	2015-10-07 + 3.33 days for every % above 98.70%

The last one at 98.70% I'm not sure of. According to the code George posted: post #1471
there is a "OR -- plus a grace period if close to finished" beyond the year and beyond the 3.33 days for every % above 10%, so it might survive longer or until it finishes, and according to Madpoo's ETA it will finish in June if left alone and if it keeps the current progress rate.

[Prime95]

Originally Posted by ATH

there is a "OR -- plus a grace period if close to finished" beyond the year and beyond the 3.33 days for every % above 10%, so it might survive longer .

You misunderstand the SQL, a danger when I posted just a SQL snippet. The SQL comment refers to the previous code on the line. The OR starts the clauses for non-grandfathered assignments that was not posted.

[cuBerBruce]

The progress on exponents 55027163, 55059383, and 55079077 appears to me to be very spiky. I believe these can jump a whole percentage point in a single day (certainly within 2 days). But typical daily progress is much lower. My current projection for 55027163 (based on data from Feb. 10) is that it will expire on Sept 3 at about 99.1% done (663 days old). But with the spiky nature of its progress it could easily get completed before expiring. EDIT: Correction, I get 662 days for Sept. 3, and I get Sept. 6 for the 665 day limit.