LLRnet/PRPnet rally January 3rd-10th - Page 4

AMDave · 2011-01-08, 08:48

The permissions are all ok.
It doesn't get that far.
the prpnet client isn't getting candidates from the prpnet server.
They are communicating, but the the client 'recv' method isn't getting the data from the server.
I already checked the firewall settings for the client exe's, but since it is already communicating that wasn't going to be the issue either.
I'd already checked the configs too.
It's a win7_64 box and the 64-bit components are uncommented in the ini.
It should be working.
The win7_32 box next to it is plugging away happily.
So is the winXP_32 box.
Running as admin makes no difference.
It seems to be the client/server interaction.
No matter. It's busy on something else now.

gd_barnes · 2011-01-08, 09:20

Checking it now...

This sounds like the dreaded memory-leak bug has not been fixed. We have intentionally not stopped-restarted the PRPnet server assuming that it was fixed.

I'll let you know.

gd_barnes · 2011-01-08, 09:33

I didn't see anything in the scrolling output that showed "too many connections" but it could have happened. Max, can you check the log file in the morning for some sort of issue? I renamed it to a "20110108" suffix, compressed it, and put it in the log folder.

I see that Lennart's output dropped by about 60 pairs in the last hour (thru 3 AM CST) but no one else's output dropped enough that I could tell if there was a problem. I'd only know by looking at the 4AM output if I stayed awake another 45 mins or so but I need to go to bed.

Regardless, I stopped and restarted the server to hopefully resolve the issue.

If it turns out there was more than a half hour of lost time, we'll add an hour to the PRPnet side of the rally.

gd_barnes · 2011-01-08, 09:33

Quote:

Originally Posted by PCZ

Dave check the permissions on the llrnet binary.
You can get that error if it isn't executable.

He's running PRPnet.

AMDave · 2011-01-08, 10:48

Gary, I don't believe it is server-side.
I waited some more and observed that all the other clients are running normally.

I just tried a fresh install of the client again with the same result.
FTR: prpclient-4.0.4beta-windows

The problem must be with my win7_64 client config. I have been fiddling some more but I still have not worked it out. Calling time on that for tonight.

Shame though. I lost the PSU on my quad in the thunderstorm that caused the blackout last night.
So I'm down 10 cores for this effort. (besides being rather un-fashionably late :P )
12 will have to do.

Dam...uh...Drat!?
the box just rebooted without warning, no BSOD.
Definitely leaving it alone for now.

rogue · 2011-01-08, 14:34

Quote:

Originally Posted by AMDave

got this from the debug log:
"[2011-01-08 18:25:28 EAST] 544: select indicates socket has data, but recv returned nothing"
Looks like GB-9000 is holding out on dn08.

That's 6 cores that are now doing something else. :(

The message is not particularly accurate. It actually implies that that server closed the socket on its end.

AMDave · 2011-01-08, 15:35

I agree.
The implication is not correct in this instance.
The problem appears to be on the client end.
Unfortunately, I cannot see why.
Should I go to debug level 2?

rogue · 2011-01-08, 17:22

Quote:

Originally Posted by AMDave

I agree.
The implication is not correct in this instance.
The problem appears to be on the client end.
Unfortunately, I cannot see why.
Should I go to debug level 2?

Typically this is a temporary issue. If it is persistent, then I would suspect that the problem is the server side.

mdettweiler · 2011-01-08, 17:34

Quote:

Originally Posted by rogue

The message is not particularly accurate. It actually implies that that server closed the socket on its end.

Indeed...I see this message show up frequently on my clients during high-load periods like rallies but it seems to be benign, as I've only ever observed it cut off the greeting message (at the very end of the communication).

@Dave and Gary: I'm not exactly sure what happened there. I suspect that at this point, looking through the logs would be a rather futile exercise since there is so much traffic in there and it would be hard to pick out Dave's problem client's attempts to communicate with the server.

Dave, I would go ahead and try debug level 2--that gives a fair amount more information that should be helpful in this case. I have the server set on debug level 1, so it's recording pretty much everything that goes by. Once you've got a debug log from your client, if you post it here I can try cross-referencing the timestamps with the server log and we can see both ends of the conversation.

Also, you may want to try running LLRnet on the problem box until we can figure out the source of this issue--that way it will at least be in the rally. The latest clients for each are equally fast so there's no particular difference except personal preference.

AMDave · 2011-01-09, 03:56

Fair enough.
Added 5 on LLRnet for a total of 17

Better leave one core for the gamer in the house

gd_barnes · 2011-01-11, 02:51

The rally officially ended a little over 6 hours ago. As expected, Brian (PCZ) took the individual title over Lennart after narrowly losing to him in the last rally and AMD users killed everyone.

The most interesting battle came down to 2nd place in the team competition. ROLP overtook PST with just 3 hours to go to win by < 900 pairs! This was the 2nd straight rally that ROLP beat PST after losing the previous 2 rallies to them. At only a 1.9% difference, this is the 2nd closest, on a percentage basis, that 2 teams have finished in the competition for any place. The June 2010 rally had PST beating ROLP for 1st place by a mere 1.3%.

Thanks to everyone who participated in our 2nd most successful rally ever on a pairs processed basis and most successful in nearly 2 years!

We found an outstanding 25 primes for n>890K!

Gary

2011-01-08, 10:48	#38
AMDave Jan 2006 deep in a while-loop 2×7×47 Posts	Gary, I don't believe it is server-side. I waited some more and observed that all the other clients are running normally. I just tried a fresh install of the client again with the same result. FTR: prpclient-4.0.4beta-windows The problem must be with my win7_64 client config. I have been fiddling some more but I still have not worked it out. Calling time on that for tonight. Shame though. I lost the PSU on my quad in the thunderstorm that caused the blackout last night. So I'm down 10 cores for this effort. (besides being rather un-fashionably late :P ) 12 will have to do. Dam...uh...Drat!? the box just rebooted without warning, no BSOD. Definitely leaving it alone for now. Last fiddled with by AMDave on 2011-01-08 at 11:07

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
LLRnet/PRPnet rally April 4th-11th	mdettweiler	No Prime Left Behind	55	2011-04-25 09:35
TPS Discovery Rally: January 14-16	Oddball	Twin Prime Search	27	2011-01-17 07:00
LLRnet/PRPnet rally Oct. 27th-Nov. 3rd	mdettweiler	No Prime Left Behind	33	2010-12-24 19:16
LLRnet/PRPnet rally August 12th-19th	mdettweiler	No Prime Left Behind	88	2010-09-09 12:50
LLRnet/PRPnet rally June 4th-6th	gd_barnes	No Prime Left Behind	61	2010-07-30 17:28

2011-01-08, 08:48	#34
AMDave Jan 2006 deep in a while-loop 2×7×47 Posts	The permissions are all ok. It doesn't get that far. the prpnet client isn't getting candidates from the prpnet server. They are communicating, but the the client 'recv' method isn't getting the data from the server. I already checked the firewall settings for the client exe's, but since it is already communicating that wasn't going to be the issue either. I'd already checked the configs too. It's a win7_64 box and the 64-bit components are uncommented in the ini. It should be working. The win7_32 box next to it is plugging away happily. So is the winXP_32 box. Running as admin makes no difference. It seems to be the client/server interaction. No matter. It's busy on something else now.

2011-01-08, 09:20	#35
gd_barnes May 2007 Kansas; USA 3³×5×7×11 Posts	Checking it now... This sounds like the dreaded memory-leak bug has not been fixed. We have intentionally not stopped-restarted the PRPnet server assuming that it was fixed. I'll let you know.

2011-01-08, 09:33	#36
gd_barnes May 2007 Kansas; USA 3³×5×7×11 Posts	I didn't see anything in the scrolling output that showed "too many connections" but it could have happened. Max, can you check the log file in the morning for some sort of issue? I renamed it to a "20110108" suffix, compressed it, and put it in the log folder. I see that Lennart's output dropped by about 60 pairs in the last hour (thru 3 AM CST) but no one else's output dropped enough that I could tell if there was a problem. I'd only know by looking at the 4AM output if I stayed awake another 45 mins or so but I need to go to bed. Regardless, I stopped and restarted the server to hopefully resolve the issue. If it turns out there was more than a half hour of lost time, we'll add an hour to the PRPnet side of the rally.

2011-01-08, 15:35	#40
AMDave Jan 2006 deep in a while-loop 1222₈ Posts	I agree. The implication is not correct in this instance. The problem appears to be on the client end. Unfortunately, I cannot see why. Should I go to debug level 2?

2011-01-09, 03:56	#43
AMDave Jan 2006 deep in a while-loop 658₁₀ Posts	Fair enough. Added 5 on LLRnet for a total of 17 Better leave one core for the gamer in the house

2011-01-11, 02:51	#44
gd_barnes May 2007 Kansas; USA 3³×5×7×11 Posts	The rally officially ended a little over 6 hours ago. As expected, Brian (PCZ) took the individual title over Lennart after narrowly losing to him in the last rally and AMD users killed everyone. The most interesting battle came down to 2nd place in the team competition. ROLP overtook PST with just 3 hours to go to win by < 900 pairs! This was the 2nd straight rally that ROLP beat PST after losing the previous 2 rallies to them. At only a 1.9% difference, this is the 2nd closest, on a percentage basis, that 2 teams have finished in the competition for any place. The June 2010 rally had PST beating ROLP for 1st place by a mere 1.3%. Thanks to everyone who participated in our 2nd most successful rally ever on a pairs processed basis and most successful in nearly 2 years! We found an outstanding 25 primes for n>890K! Gary