24 hour rally

[gd_barnes]

Originally Posted by em99010pepe

I'm going to clear the cache from 5 cores then move to manual reservation. I think the best strategy is to manual reserve ranges when possible and only add cores to LLRnet in limit cases like work machines where you don't have daily control, etc. I prefer to use the manual client so from now on I won't run LLRnet client.

I need to know who can only run LLRnet?
My example: only on 2 cores, the rest can run manual LLR.

We need to soft the interaction between clients/server until we reach a higher n when the candidates began to be longer to test and therefore less communications with the server will take place.

I just now woke and saw everything since 8 AM GMT.

I have to defer to Carlos here. If he thinks it is best to call off the rally, then that is what we will do.

Carlos, the only alternative that I might suggest: If everyone put their LLRnet machines on port 300 and cancelled any work on all other ports including the CRUS ports, would that make the rally possible?

I can run all of my machines on manual LLR as I've been doing all along. I have 3 work machines running manual LLR on the NPLB high-range right now because I felt it too risky too transfer them all over to the LLRnet server at this time.

I'll have to make a run into work in the next 1-2 hours to get them changed over.

The time-consuming thing for me is dividing up and posting a bunch of files for everyone to reserve on both the lower and higher ranges.

Carlos, I'll check again in 45 mins. after I shower/eat. If the rally is off, I'll start posting files for the high range and see what the low range looks like at that time.

If I start posting manual files to reserve, should I assume that LLRnet is still keeping the same reservations? I will assume that LLRnet will not take n=372K-392K on drive 1 since it has not started yet. I'll start by dividing up that range if the rally is off.


Gary

[mdettweiler]

A-ha! I think I've figured out what the cause of the problem is!

If you look in llr-serverconfig.txt near the top you see the following lines:

-- maximum number of concurrent connections
maxConnections = 8

Carlos, what is maxConnections set to on your LLRnet servers? If it's set to the default of 8 as shown above, then how about you try bumping it up to, say, 15?

[em99010pepe]

Originally Posted by Anonymous

A-ha! I think I've figured out what the cause of the problem is!

If you look in llr-serverconfig.txt near the top you see the following lines:

-- maximum number of concurrent connections
maxConnections = 8

Carlos, what is maxConnections set to on your LLRnet servers? If it's set to the default of 8 as shown above, then how about you try bumping it up to, say, 15?

I had 16 on server port 300 and only 8 one server port 100. Now all are set to 30.

[mdettweiler]

Originally Posted by em99010pepe

I had 16 on server port 300 and only 8 one server port 100. Now all are set to 30.

Okay. Both my clients seem to be doing fine now (they're both on port 300 at the moment)--hopefully that fixed it!

[em99010pepe]

Gary,

I have my doubts about the rally. Since my morning that I've been adjusting the servers settings, reading the RieselSieve forum to sort what's going on and observing the server behave and every 2-3 hours the server stops reacting. With last modification, thanks Anon, I hope we all manage to fix this problem.

I'll be awake for the next 6-7 hours and then how can I reset the servers? I rather sleep than stay awake to control things. Anyone of you guys use LogMeIn to remotely control your machines?

Server port 300 is giving pairs in n=365.5k range of max of 370k so I suppose it's better to keep the reservation of n=372k-392k.

[gd_barnes]

Originally Posted by em99010pepe

Gary,

I have my doubts about the rally. Since my morning that I've been adjusting the servers settings, reading the RieselSieve forum to sort what's going on and observing the server behave and every 2-3 hours the server stops reacting. With last modification, thanks Anon, I hope we all manage to fix this problem.

I'll be awake for the next 6-7 hours and then how can I reset the servers? I rather sleep than stay awake to control things. Anyone of you guys use LogMeIn to remotely control your machines?

Server port 300 is giving pairs in n=365.5k range of max of 370k so I suppose it's better to keep the reservation of n=372k-392k.

Is 30 connections enough? I will 11-12 cores running myself. Does 1 connection = 1 core? Should it be 100-150 connections? Or perhaps it only needs enough connections as there are # of people trying to access it at any one moment, which would be a small percentage of the total cores running.

I say the rally is ON! Let's get it started at 7 PM GMT. If things go haywire in a few hours, we can call it off then. It doesn't hurt to start it and then stop it. It's possible that the # of connections fixed the issue.

Can everyone switch their machines to port 300 for the rally? Short of time now...I gotta eat, will check the response here in 20 mins, then run into work to change everything back to port 300.

Another option would be to delay the start to 8 or 9PM GMT to allow people time to get all their machines changed to port 300.

What does everyone think? Can all the Free-DC folks get all their machines changed over to port 300 in the next hour? If not, could they do that by 8 or 9 PM GMT?


Gary

[mdettweiler]

Originally Posted by gd_barnes

Is 30 connections enough? I will 11-12 cores running myself. Does 1 connection = 1 core? Should it be 100-150 connections? Or perhaps it only needs enough connections as there are # of people trying to access it at any one moment, which would be a small percentage of the total cores running.

Connections means the number of clients that can actually talk to the server at a time--i.e. grabbing a new k/n pair or reporting one complete. So 30 connections should be plenty.

I say the rally is ON! Let's get it started at 7 PM GMT. If things go haywire in a few hours, we can call it off then. It doesn't hurt to start it and then stop it. It's possible that the # of connections fixed the issue.

Can everyone switch their machines to port 300 for the rally? Short of time now...I gotta eat, will check the response here in 20 mins, then run into work to change everything back to port 300.

Another option would be to delay the start to 8 or 9PM GMT to allow people time to get all their machines changed to port 300.

What does everyone think? Can all the Free-DC folks get all their machines changed over to port 300 in the next hour? If not, could they do that by 8 or 9 PM GMT?


Gary

I think Carlos said that because of the server problems, the rally will include both port 300 and port 100. So I guess we don't need to do any switching at all (now that both 300 and 100 are set to 30 connections each and should be able to handle the load).

[em99010pepe]

My last position: I prefer the rally on port 100 to clean the lower ranges but I have doubts if the server can keep the pace with all power involved in only one port so it's better to keep the decision to each one of you. If you want to find some Top 5000 primes you run on port 300, if you want to help to clean Gary's mess you should run on port 100. lol

[mdettweiler]

Originally Posted by em99010pepe

My last position: I prefer the rally on port 100 to clean the lower ranges but I have doubts if the server can keep the pace with all power involved in only one port so it's better to keep the decision to each one of you. If you want to find some Top 5000 primes you run on port 300, if you want to help to clean Gary's mess you should run on port 100. lol

Hmm...what do you mean by "Gary's mess"? I don't quite get your drift there.

[em99010pepe]

Originally Posted by Anonymous

Hmm...what do you mean by "Gary's mess"? I don't quite get your drift there.

I'm teasing him...
If we decide to run the rally on both ports Kar_born will have to make two race stats.

[gd_barnes]

Originally Posted by em99010pepe

I'm teasing him...
If we decide to run the rally on both ports Kar_born will have to make two race stats.

EXCELLENT!! OK, here's what I will do...I'll leave my work machines on port 100; that's 7 total. I'll run my home machines on port 300; 4 Athlon cores and my 3.2 Ghz P4; 5-6 cores total. I'm getting ready to start the rest of the home machines now.

Carlos, can you do me a favor? Can you see if I'm getting some major processing on port 100 right now? With 7 machines on it, I should be sending back about 2+ results per minute to the server. If you can do that, it'd save me a trip into work to check on things.

Thanks Carlos and Anon for all of your hard work in getting the problems diagnosed.

For the time being, I won't split up any files until I hear otherwise.

EDIT: I'm perfectly fine with running all 13 cores on port 100 but I don't want to risk overloading one server. Technically I'm giving up some score in the rally by doing so since the higher range takes longer to LLR. At this point, it's all about what makes the rally run smoothest...nothing else is as important as that.


Thanks,
Gary