LLRnet servers for NPLB

[Mini-Geek]

Originally Posted by gd_barnes

I had no programmatic means to remove the headers and the files were too big to use Excel formulas and sorting. It was taking me 10-15 mins. per file to remove them manually so I stopped after the first two but sent all 5 files.

Three words:
Find and Replace

Only issue is that you need to remove the newline character after the extra headers. Depending on your text editor's find format, this might not be possible. Notepad++ lets you choose a mode (Extended) where \n is understood to be a new line. Metapad also understands \n as newline, but doesn't handle very large (>30MB or so) files as well as other editors.

[gd_barnes]

Originally Posted by Mini-Geek

Three words:
Find and Replace

Only issue is that you need to remove the newline character after the extra headers. Depending on your text editor's find format, this might not be possible. Notepad++ lets you choose a mode (Extended) where \n is understood to be a new line. Metapad also understands \n as newline, but doesn't handle very large (>30MB or so) files as well as other editors.

At the time that I sent the files, I had no means because all I had was "regular" notepad and Excel 2003 (65536 lines max). Now I'm able to do it fairly readily and quickly. Thanks for the tip.

[mdettweiler]

Success! The tweaked version of prpserver now hands out pairs by k primary and n secondary, and I've verified that this does not affect how the webpage or prpserver.candidates are handled.

One thing that helps is that PRPnet keeps everything insulated quite nicely. The order in which pairs are handed out, the order in which they're displayed on the webpage, and the order in which they're written to prpserver.candidates are all defined in distinct pieces of code, meaning that they can be tweaked without affecting any of the others.

I'll post an announcement in the PRPnet thread shortly about the new G7000's now being online.

Edit: Oh, and sometime later today I'll see about the tweak to have prpserver.candidates sorted by n. That should be rather easy.

[gd_barnes]

Max,

I'm still seeing problems on PRPnet server port 7000. It's still getting the "nothing was received, therefore the socket was cloed" message. (Yes, in the message "closed" is spelled "cloed". Perhaps that could be changed. lol) I'm sure it's due to load but we've been down this road before. Tweaks won't fix these kinds of load issues. Based on competent technical advice that I have received, I believe that PRPnet still uses memory for too many things.

I received word that PrimeGrid is struggling with PRPnet. PRPnet is still not ready yet to roll completely out yet at NPLB. We may end up having to go back to an LLRnet server for the 12th drive. We'll try to muddle our way through the k=2400-2600 range but I don't want to keep fighting it.

In the mean time, the 6th drive LLRnet server will likely dry within a week. When it is convienient over the next few days, please create a new LLRnet server on my machine for the next range. For the time being, that is what we will do to migrate servers off of Ironbits machine...just create a new LLRnet server on my machine as the servers are near drying.

For now, we'll just leave the 5th and 12th drives and doublecheck effort on PRPnet servers with the strong possibility of going back to an LLRnet server on the 12th drive. Even though the higher n-ranges shouldn't have a PRPnet problem due to longer tests, I don't want to chance it.

Bruce and Lennart, for now, you might consider the 5th drive port G4000 PRPnet server. It can handle the load because the tests are much bigger and so don't process so quickly.


Gary

[mdettweiler]

Originally Posted by gd_barnes

Max,

I'm still seeing problems on PRPnet server port 7000. It's still getting the "nothing was received, therefore the socket was cloed" message. (Yes, in the message "closed" is spelled "cloed". Perhaps that could be changed. lol) I'm sure it's due to load but we've been down this road before. Tweaks won't fix these kinds of load issues. Based on competent technical advice that I have received, I believe that PRPnet still uses memory for too many things.

I received word that PrimeGrid is struggling with PRPnet. PRPnet is still not ready yet to roll completely out yet at NPLB. We may end up having to go back to an LLRnet server for the 12th drive. We'll try to muddle our way through the k=2400-2600 range but I don't want to keep fighting it.

In the mean time, the 6th drive LLRnet server will likely dry within a week. When it is convienient over the next few days, please create a new LLRnet server on my machine for the next range. For the time being, that is what we will do to migrate servers off of Ironbits machine...just create a new LLRnet server on my machine as the servers are near drying.

For now, we'll just leave the 5th and 12th drives and doublecheck effort on PRPnet servers with the strong possibility of going back to an LLRnet server on the 12th drive. Even though the higher n-ranges shouldn't have a PRPnet problem due to longer tests, I don't want to chance it.

Bruce and Lennart, for now, you might consider the 5th drive port G4000 PRPnet server. It can handle the load because the tests are much bigger and so don't process so quickly.


Gary

Hey Gary, I think this new patch will fix the problem. I haven't gotten the chance to apply it yet, and as such that may be why we're still seeing those problems.

Aside from this little problem, this latest version of PRPnet has been working exceedingly well for us. I'd like to give it a bit more of a chance unless any other distinct issues arise.

[gd_barnes]

OK, let's watch it closely for 2-3 days, mainly during the times when both Bruce and Lennart are on there with many cores. If PRPnet can survive those 2 on small n-ranges for a day or more, I'll at least be confident enough to go ahead and keep the 12th drive on it and possibly create one for the 6th drive within the next week as the current LLRnet server dries up.

[Brucifer]

Okay, I put some systems back on 7000.

[gd_barnes]

LLRnet port 5000 for the 6th drive has now nominally dried out. That is, it's not sending out any more new work. For continuing on that drive, see new PRPnet port 5000. Details are in the "PRPnet servers for NPLB" thread.

[gd_barnes]

The Email notification has stopped working on port IB2000. I had 2 primes from ~3 days ago that I just now saw in the primes.txt file.

[AMDave]

The log file "nplb_hourly.sh.20091205170001.log" shows that the email was sent

Starting mail_notification at Sat Dec 5 17:00:43 2009
Debugging is off
0 seconds in total elapsed...
About to process mail notifications...
Updating notification exceptions...
Processing notifications...
Sent notification for gd_barnes for 2769 * 2 ^ 512325 - 1 ...
Updating mail sent flags...
31 seconds in total elapsed...
Processed 2 sql statements...
Completed mail_notification at Sat Dec 5 17:01:14 2009

no error was recorded in the compilation or sending of the email and no email rejection was received by the server from upstream.

Do you have a spam / junk folder?

[gd_barnes]

Originally Posted by AMDave

The log file "nplb_hourly.sh.20091205170001.log" shows that the email was sent

no error was recorded in the compilation or sending of the email and no email rejection was received by the server from upstream.

Do you have a spam / junk folder?

Yes I do and I checked that. Nothing there.

The above is not one of the primes that had a problem. The two primes that I did not get were on Dec. 9th:

2193 513360 gd_barnes
[2009-12-09 14:35:33]
2193*2^513360-1 is prime! Time : 2250.0 sec.

2967 513388 gd_barnes
[2009-12-09 17:32:11]
2967*2^513388-1 is prime! Time : 2285.0 sec.

Gary