LLRnet and PRPnet servers for automated LLR

[mdettweiler]

Originally Posted by mdettweiler

Okay, thanks for the heads-up. I need to tweak my LLRnet-format conversion script to handle the different output format used for supplemental tests.

Fixed now. Should be OK for all future such supplemental tests. (Well, there is a small chance that it might barf on an actual twin since I wasn't able to test it on actual data from such an occurrence...but I'll cross that bridge if and when we come to it.)

[Oddball]

The LLRnet server has dried out.

Max, could you load n=486001-486300 into port 12050? Usually, I would ask for a bigger range (n=486001-487000), but I'm waiting until April's sieve rally is over before loading n=486301 and up.

[mdettweiler]

Originally Posted by Oddball

The LLRnet server has dried out.

Max, could you load n=486001-486300 into port 12050? Usually, I would ask for a bigger range (n=486001-487000), but I'm waiting until April's sieve rally is over before loading n=486301 and up.

Okay, I'll do that a little later today.

[Oddball]

Max, could you load 100K<k<200K, n=480000-481000 into the PRPnet server and k<100K, n=486300-487000 into the LLRnet server? Port 12000 and 12050 have both dried out.

Use the newest sieve files, which have been posted by gribozavr here:

http://www.mersenneforum.org/showpos...26&postcount=8

[Lennart]

http://prpnet.mine.nu:12098/all.html

Here you can see a test I have done on Twin search.
This is on PRPNet 4.3.2

As you can see I added n= 333333 twin just to test that it works.

You also get all composite +1 test with residue on those -1 test that is prime.

+1 test will only be done if -1 test is Prime

It start test -1 ( if you load -1) then +1 and if that also is prime it does a GF test.

Lennart

[Oddball]

It looks like there's a mini-rally going on now Today, 18217 k/n pairs were tested - 16458 on PRPnet and 1759 on LLRnet. This is TPS's second highest daily LLR rate, trailing only the 29,961 pairs tested on January 16th, which was the last day of the TPS Discovery Rally.

As you might have guessed, this means that the PRPnet server will dry out soon. So Max, could you load n=481000-482000 into the PRPnet server? There are only ~2500 candidates in port 12000, and that number will hit zero in a couple of hours at the rate we're going now.

[ellipse]

Originally Posted by Oddball

It looks like there's a mini-rally going on now

The "mini-rally" may be turning into something big. 1316 pairs were LLR'ed in the past hour, which will be a record 31,584 pairs/day if it's sustained.

Anyone up for an impromptu rally?

[mdettweiler]

Originally Posted by Oddball

It looks like there's a mini-rally going on now Today, 18217 k/n pairs were tested - 16458 on PRPnet and 1759 on LLRnet. This is TPS's second highest daily LLR rate, trailing only the 29,961 pairs tested on January 16th, which was the last day of the TPS Discovery Rally.

As you might have guessed, this means that the PRPnet server will dry out soon. So Max, could you load n=481000-482000 into the PRPnet server? There are only ~2500 candidates in port 12000, and that number will hit zero in a couple of hours at the rate we're going now.

Looks like Lennart decided to give his newly expanded farm a whirl here.

My apologies if any record-breakings were prevented by the inconvenient drying of the server...it's reloaded now. I'll continue to keep a close eye on it as long as this period of high activity continues.

@ellipse: interesting idea--how high can we go? I can't join in myself at the moment (leaving for church in a few minutes and don't have time to switch anything over) but I may be able to finagle something when I get home.

[ellipse]

Originally Posted by mdettweiler

@ellipse: interesting idea--how high can we go? I can't join in myself at the moment (leaving for church in a few minutes and don't have time to switch anything over) but I may be able to finagle something when I get home.

Right now (15:30 CDT), we're at 9883 tests. If we continue at a rate of 1200 tests/hour or more for the rest of the day, we'll get above 20,000 tests. That's higher than yesterday's rate but lower than the all-time rate on January 16th.

What we can do is set a new 3-day record and a new 24 hour record tomorrow. The Jan 14-16 discovery rally had 53,179 pairs tested at an average of 17,726 pairs/day. That's lower than yesterday's rate and likely lower than today's rate as well. If we break the 24 hour record tomorrow, we'll be looking at ~70,000 pairs completed from April 23-25: ~18,000 pairs yesterday, ~20,000 pairs today, and possibly ~32,000 pairs tomorrow.

My internet connection is a bit spotty at the moment, but I'll put some cores on this as soon as the problem is resolved.

[Oddball]

The party's over We finished 19,119 pairs yesterday (16989 on PRPnet and 2130 on LLRnet), which was higher than the day before but not enough to beat January's record. Today's halfway over, and we're only at 4135 candidates tested. That's a higher rate than usual, but far short of what's needed to set a new record. I'll be pulling off my cores soon to work on sieving and to verify the primes that have been found during the past few days.

One more thing: Max, could you remove the rows for n=480000-481500 and n=483001-484000 on the PRPnet stats page? All k/n pairs in those ranges have been tested, and it'll make this page load much faster:
http://noprimeleftbehind.net:12000/all.html

Finally, I'd like to welcome Beyond to TPS. He put a core or two on the LLRnet server yesterday.

[mdettweiler]

Originally Posted by Oddball

One more thing: Max, could you remove the rows for n=480000-481500 and n=483001-484000 on the PRPnet stats page? All k/n pairs in those ranges have been tested, and it'll make this page load much faster:
http://noprimeleftbehind.net:12000/all.html

Okay, I've removed n=480000-480999 and n-483001-484000. The way the database is set up, I need to match ranges based on text strings (for instance, "%*2^480%-1" where % is any character) rather than on integer ranges; thus, it's somewhat inconvenient to remove chunks more granularly than n=1K at a time.