OFFICIAL "SERVER PROBLEMS" THREAD

[Uncwilly]

Originally Posted by drkirkby

For example, one of my entries is
PRP=xxxxxxx,1,2,104760107,-1,76,2,3,1
which, if my understanding is correct, will result in the P-1 test taking longer than the optimal amount.

There is nothing preventing you from changing the value to 1.1 tests saved versus 2. You could even write a chron job that hourly checks for new assignments and fixes the entries in the worktodo. kriesel explained why PrimeNet might not want to change over from 2 tests saved straight to 1. Maybe 6 months after April 8 (the end of first first time LL's being handed out) the value can start be set to 1.5. Then if the data support it, later to 1.2 or 1.1

[Madpoo]

Originally Posted by Viliam Furik

If you mean Aaron (user Madpoo), he's basically a magician - he's in charge of the PrimeNet server. Therefore it's easy for him to make a direct search in the database. But even a regular PrimeNet user can do it, by carefully looking through detailed exponent status for the exponent range 103M - 104M. With enough time on your hands, you can check which exponents are not assigned. 54206 exponents are not that much, considering most of them are factored...

It's not really magic - it's more like cheating. When you have access to query the database directly, it's nearly trivial.

When it comes to "interesting" requests for data that:
a) isn't readily available from the website data (or would be intensive to get that way)
b) is something that would be useful to more than just a few people

We're amenable to those requests - we've generated reports and posted the data here on the forum.

Fact is, we'd rather do a simple SQL query and spit out the data than have someone tying up the server with millions of web requests for days at a time. LOL

And if it's something that we get regular requests for, we may even bake it into an existing report, or a brand new one. I think James did that recently for a request to spit out lists of all unassigned exponents that have mismatching LL results that need a 3rd (or more) test to resolve.

I used to spit those lists out more regularly back when there was a huge backlog, but now there aren't quite as many.

[drkirkby]

Is the data on milestones for first-time checks at https://www.mersenne.org/report_milestones/ accurate?

I was looking at the data at https://www.mersenne.org/report_milestones/ to see if I could estimate where first time checks would be in 6 months time. I'm a bit puzzled by the data which shows the date of first time tests of exponents of different values. According to that webpage, it has taken

* Approximately 6 months to get the first time tests from 90 million to 92 million
* Approximately 1 month to get the first time tests from 95 million to 100 million
* Approximately 6 months to get the first time tests from 100 million to 103 million
* 97 & 98 million were completed on the same day (2020-11-17).

Are these really correct, or are there some errors? I find it a bit hard to believe it can take 6 months to cover 1 million, then 5 million are completed within one month. Here's a copy of the data from the webpage

Date ...........First time checks
2021-06-05 All exponents below 103 million tested at least once.
2021-04-11 All exponents below 102 million tested at least once.
2021-01-15 All exponents below 101 million tested at least once.
2020-12-04 All exponents below 100 million tested at least once.
2020-11-27 All exponents below 99 million tested at least once.
2020-11-17 All exponents below 98 million tested at least once.
2020-11-17 All exponents below 97 million tested at least once.
2020-11-09 All exponents below 96 million tested at least once.
2020-11-03 All exponents below 95 million tested at least once.
2020-10-24 All exponents below 94 million tested at least once.
2020-10-09 All exponents below 93 million tested at least once.
2020-10-04 All exponents below 92 million tested at least once.
2020-06-20 All exponents below 91 million tested at least once.
2020-04-09 All exponents below 90 million tested at least once.

[Uncwilly]

Those are the dates for when the last numbers were completed. Not the bulk of the numbers being completed. Stragglers are the laggards. That does not reflect the number of exponents per day that are being completed. You are looking at the wrong metric.

[kriesel]

As it happens, it can be seen as in https://www.mersenne.org/assignments...chk=1&excert=1
103580003
103832639
are all that currently remain in 103M-104M. Both appear likely to expire and be reissued, increasing the time for completion of the 104M milestone somewhat.

[drkirkby]

Originally Posted by Uncwilly

Those are the dates for when the last numbers were completed. Not the bulk of the numbers being completed. Stragglers are the laggards. That does not reflect the number of exponents per day that are being completed. You are looking at the wrong metric.

Thank you. That makes sense, although I'm surprised the laggards did not get their exponents expired by the exponent moving midway into the first category, at which point it should be unassigned, and someone else complete it quickly
Given I am currently receiving exponents around 104.5 million, is there any reasonable way to estimate what the exponents will be in 6 months time? (I was interested in this with a view to using the information to determine at what FFT values were worth benchmarking.)

[Viliam Furik]

Originally Posted by drkirkby

Thank you. That makes sense, although I'm surprised the laggards did not get their exponents expired by the exponent moving midway into the first category, at which point it should be unassigned, and someone else complete it quickly
Given I am currently receiving exponents around 104.5 million, is there any reasonable way to estimate what the exponents will be in 6 months time? (I was interested in this with a view to using the information to determine at what FFT values were worth benchmarking.)

106.5M-107M, give or take. Please note this is a pure guesstimate based on the number of non-tested assignments.

The corresponding FFT size for the aforementioned range is 5760K. 104.5M should have 5600K in Prime95.

The milestones are that way because the PrimeNet work assigning is not sequential, but rather categorical (assignment categories and rules), thus spread out further forward. That way it can happen that the 98M range is already complete, but 97M is not; therefore for both of them to be complete, the 97M must be completed.

[slandrum]

Originally Posted by Viliam Furik

The milestones are that way because the PrimeNet work assigning is not sequential, but rather categorical (assignment categories and rules), thus spread out further forward. That way it can happen that the 98M range is already complete, but 97M is not; therefore for both of them to be complete, the 97M must be completed.

In addition, there was a massive push to get over the 100M milestone, and Ben Delo had a lot more computing power working on FTC than is running right now, so a lot of milestones were closing at the same time.

[kriesel]

Originally Posted by drkirkby

is there any reasonable way to estimate what the exponents will be in 6 months time?

Yes. Use the reference info. Bookmark it. Keyword search it. Skim it. Read some, think, and understand. There's a whole stats section. A GIMPS progress subsection. Including periodically recorded snapshots of where the wavefronts of various computation types lay on specified dates. Extrapolate.
Projections made in December 2018 with stated assumptions, versus observed history.

[S485122]

First of all the LL results query (https://www.mersenne.org/report_ll/) works as if the upper limit of the range to be queried defaults to 999999937. This means that just going to that page implies a query is started that stops to display the first 10000 results and after that if one does only specifies the lower bound (to query the LL results for a particular exponent for instance) again one gets the first 10000 results and the upper bound changes to 999999937. Other queries default to an upper bound equal to the lower bound and thus avoid trashing the database whenever the page is opened.

Then the database has a field for the status of LL and PRP tests. This field allows the residue "78BA" to be considered verified for exponent 1277. The current query doesn't show the status of the tests for factored Mersenne numbers. They are grouped in the "Later Factored test results" category. But the LL (or PRP) test still is verified, unverified or bad (it certainly hasn't been factored ;-). If that information isn't deleted for all test results when a factor is found, it would be great to be able to retrieve it.

Jacob

[kriesel]

www.mersenne.org or pages beneath get the hourglass, not the page. No manual assignments or reporting.