LLRnet servers for NPLB

[gd_barnes]

Originally Posted by IronBits

You dog! You stole the 1st prime to, with only one quad?!?!!? /kicks can

Huh, you think that is bad...Tnueral found a NEW n>600 prime on port 5000 with a single slow core on it vs. my 16 fast cores! lol /kicks TWO cans!

His core is so slow that it processes 2-3 results per hour. lmao I guess he had to get one sometime.


Gary

[IronBits]

Ya, I saw that! /bangs head
henryzz auto-notify email just went out because he found one just now

[IronBits]

I need more work for port 8000 asap!

[mdettweiler]

Originally Posted by gd_barnes

David,

I noticed that Nuggetprime is dumping huge #'s of results on port 8000 all at once every few hours. The most reacent; 998 in one hour was the largest.

Is this creating any problems? I thought we were having serious issues when Lennart returned many pairs at once like that.

Nugget, thanks for the huge contribution to the effort! I realize it is slightly faster to cache as many as you can, run them manually, and then return them to the server. It is my preference if you cache just a few (< 200) and return them in smaller pieces (< 100 at a time). Caching so many at once makes it difficult for us to tell where we're at and returning so many at once has caused server problems in the past. If you cache a whole bunch and then keel over, we'll end up waiting the JobMaxTime for them to be processed. lol

David, please change the JobMaxTime on port 8000 to 1 day. Thanks.

It looks like we're going to smoke through this range in < 2 days!! 21,600 pairs done; 18,100 pairs to go! And I only have one quad on the effort!! Weeee, what a ride!


Gary

Gary, the problems we were having with LLRnet were due to *proxies*, not due to users caching k/n pairs to crunch manually. Yes, the dumps can look kinda similar, but the process is very different. That's because when you cache manual k/n pairs from the server, when you return the results to the server it is done through the LLRnet *client* software (not the proxy server software)--which doesn't hog up sockets on the server like the proxies do. Thus, there is no danger of crashing the server this way.

[mdettweiler]

Originally Posted by IronBits

I need more work for port 8000 asap!

Hmm...Gary didn't reply to my query over in the Drive #8 thread about loading work starting at 200K into the server (maybe it got lost in all the other replies there?), but since time is of the essence I'll go ahead and send you 200K-210K shortly; that should keep port 8000 going for a reasonable amount of time.

In the meantime, we should report any primes found in this range in the Drive #8 thread, though when Gary sees this he can decided if we should keep track of the lower range in a new team drive.

[IronBits]

I just want more primes!!! <lol>

I also want everyone to crash one of the 3 servers, or at least try to, without using proxies
With the last batch of work you gave 8000, seems to have garnered quite a bit of interest... I'd like to keep that up if possible

[mdettweiler]

Originally Posted by IronBits

I just want more primes!!! <lol>

I also want everyone to crash one of the 3 servers, or at least try to, without using proxies
With the last batch of work you gave 8000, seems to have garnered quite a bit of interest... I'd like to keep that up if possible

Well, these numbers are even smaller (the smallest, in fact, that our current best estimates say can be run in an LLRnet server safely)--so if anything's going to crash your server, these will.

[mdettweiler]

Originally Posted by IronBits

I have added jobMaxTime and prunePeriod to the main page, so you can how those values are set for each port.

http://nplb.ironbits.net

It shows
Port 8000 has 18124 remaining knpairs!
knpairs.txt - Size: 217512
Last Updated: 1/1/2009 11:23:33 PM
first k/n pair 1093 350215
last k/n pair 1967 352000
jobMaxTime = 1 * 24 * 3600 -- 1 days
prunePeriod = 1 * 24 * 3600 -- 1 days

Okay, cool! One thing, though: can you change prunePeriod on all three servers to 1 hour? I think having it set to such a large amount may be responsible for ranges hanging around a lot longer as "in progress" then they really need to.

[IronBits]

Roger that, I'll get right on it.
Send double the range you were going to if they are smaller/faster ;)

[mdettweiler]

Originally Posted by IronBits

Roger that, I'll get right on it.
Send double the range you were going to if they are smaller/faster ;)

That's why I sent you a 10K range this time instead of 5K like I was originally thinking of. (For comparison: the relatively bigger/slower work we have in port 8000 right now was loaded in a range of 2K.)

[IronBits]

Where is the :roflmao: icon when you need it?
prune period was fine, the variable I was using to print was not accurate ;)

Port 8000 has 207657 remaining knpairs!
knpairs.txt - Size: 2491908
Last Updated: 1/2/2009 10:32:36 AM
first k/n pair 1945 350595
last k/n pair 1955 210000
jobMaxTime = 1 * 24 * 3600 -- 1 days
prunePeriod = 1 * 1 * 3600 -- hourly