C443 server issues on 4th drive

[em99010pepe]

Originally Posted by gd_barnes

About how frequently is a k/n pair being returned?

A pair every 30 secs and with dumps from my side every 6 hours or so.

Why don't you host a server just for you?

[gd_barnes]

Originally Posted by em99010pepe

A pair every 30 secs and with dumps from my side every 6 hours or so.

I think a test would take ~475-525 secs. or so. So perhaps 500 / 30 = 16-17 cores are on it right now.

That's not really very many. Hum.

Dumps every 6 hours? I wonder if something went haywire with the last dump? If you drop the server and bring it back up, will it cause everyone on it to be dropped? If so, don't do that until more people are around.

[gd_barnes]

Originally Posted by em99010pepe

Why don't you host a server just for you?

I've tried it. I can't figure it out. It may be easy for you guys but I spent hours trying to get one to work several months ago. Don't have time right now.

Also, that's not the point. We need the main servers on this drive to work correctly for everyone.

[gd_barnes]

OK, it just started working. lol

What did you do because I didn't do anything differently on my end? It just now sent that result that had been stuck in my 'tosend' file and got a new k/n pair to test...after I hit 'llrnet.exe' twice.

[em99010pepe]

Got your pair after rebooting the server and forcing a prune. Now I am going back to bed. Bye.

Edit: Check your task manager to see if you have more than a instance of llrnet running per core.

[gd_barnes]

Originally Posted by em99010pepe

Got your pair after rebooting the server and forcing a prune. Now I am going back to bed. Bye.

I hope no one else lost their connection as a result of the reboot. Everyone, check their connections to C443 when you're on next...

Thanks for doing...

[gd_barnes]

Originally Posted by em99010pepe

Got your pair after rebooting the server and forcing a prune. Now I am going back to bed. Bye.

Edit: Check your task manager to see if you have more than a instance of llrnet running per core.

I had already done that. I was having a problem on all machines when trying to connect the 1st core of a quad with nothing else running.

My Windows machine is connected now. I'll try the quads now.

Edit: This has now been moved to a different thread.

[IronBits]

From my observations when I first started running an llrnet server under Windows, I noticed it had a limited supply of sockets per port.
The problem stems from when the a socket hangs and then the OS begins enumerating sockets until the port stops working.
Rebooting always releases the whole mess, but begins to happen again...

As I recall, it's when a remote client tries to connect but fails (sometimes over and over) and before it can release that socket, another connection comes in, then another and so on, by the time the whole sequence is finished, a socket will get locked up, eventually running out of sockets on that port number.

Rebooting or restarting the llrnet server in no way 'disconnects' clients. The clients just keep asking for work until the server comes back online.
There appears to be an amount of time the client will cease to function if it can't talk to the Server, but the time it takes to reboot a computer or restart a llrnet server, should not have any impact on the clients hanging..

I moved all my servers over to Linux (CentOS 64bit) to get away from that problem. Linux appears to handle that situation much better.
I know the llrnet server software can not handle too many cores because of the high connect/disconnect rate will cause it to fail eventually.
I do not believe it was written to handle hundreds of cores at once. The only option is to have many llrnet servers, hopefully running under Linux, probably being the better solution.

I am currently running 6 llrnet servers over here, each using a different port of course, on just one Computer and it seems to be doing great so far.
If you need/want me to run another llrnet server on port 443 to help balance the load, for those that need to use that port, I can do that to.
The llrnet Server software has very little impact on the Server resources itself.

[gd_barnes]

Originally Posted by IronBits

From my observations when I first started running an llrnet server under Windows, I noticed it had a limited supply of sockets per port.
The problem stems from when the a socket hangs and then the OS begins enumerating sockets until the port stops working.
Rebooting always releases the whole mess, but begins to happen again...

As I recall, it's when a remote client tries to connect but fails (sometimes over and over) and before it can release that socket, another connection comes in, then another and so on, by the time the whole sequence is finished, a socket will get locked up, eventually running out of sockets on that port number.

Rebooting or restarting the llrnet server in no way 'disconnects' clients. The clients just keep asking for work until the server comes back online.
There appears to be an amount of time the client will cease to function if it can't talk to the Server, but the time it takes to reboot a computer or restart a llrnet server, should not have any impact on the clients hanging..

I moved all my servers over to Linux (CentOS 64bit) to get away from that problem. Linux appears to handle that situation much better.
I know the llrnet server software can not handle too many cores because of the high connect/disconnect rate will cause it to fail eventually.
I do not believe it was written to handle hundreds of cores at once. The only option is to have many llrnet servers, hopefully running under Linux, probably being the better solution.

I am currently running 6 llrnet servers over here, each using a different port of course, on just one Computer and it seems to be doing great so far.
If you need/want me to run another llrnet server on port 443 to help balance the load, for those that need to use that port, I can do that to.
The llrnet Server software has very little impact on the Server resources itself.

Thanks David. I've observed that yours seem to be able to handle up to ~50-60 cores at once at n=~500000 but that should increase as the n-values (the 2nd number in each line of the file that we send you) get larger because the testing times increase quite a bit.

We'll see how port 443 holds up over time and if there are too many issues, we may have you set up another one for us.

I've kept my 2 quads on your port 7000 for now. The others are blowing through port 443 pretty fast as it is so my 2 quads will help balance things out.

At our current rate, I think we'll blow through n=5K-8K per day combined on the 2 servers! Looking good!


Gary

[gd_barnes]

C443 clearly has issues with too many clients.

I just now noticed that the last single k/n pair that I got from it ~2 hours ago is sitting in my 'tosend' file and once again, the results file says 'Could not send result'. The k/n pair is 175*2^601991-1.

I tried starting it again twice...no luck.

David, hold off on this until you get the official word on this but: I think we're going to need to have you set up a 2nd server for this drive. We should probably run C443 dry and then have everyone move over their machines.

Carlos, is there anything you can think of to permanently fix this problem? Maybe run it on Linux like suggested by David? If not, we should probably have David running the 2 main servers for the drive.

You could still keep your C443 server intact for you and/or your buds as long as you kept it at perhaps 16 cores or less at all times. Everyone else such as Anon, me, etc. should probably use David's 2 servers, assuming that we go that route. In effect, it'd be just like our 1st drive is now.


Gary

[em99010pepe]

Originally Posted by gd_barnes

Carlos, is there anything you can think of to permanently fix this problem? Maybe run it on Linux like suggested by David? If not, we should probably have David running the 2 main servers for the drive.

I have here a machine with Ubuntu but I cannot turn it on due to energy savings. When I get the time I'll check for a few tweaks like opening more connections, etc.
I'll keep the server for Jokern...

Carlos