LLRnet servers needed

[IronBits]

I was able to handle 75 cores of Beyond + (the octo with no proxy).
You guys just tell me what to put on which port and I'll get er done.

[mdettweiler]

Originally Posted by gd_barnes

Maybe we can use this server for 300<k<400 for n>333.3K in 3 days after I'm done sieving and have the files ready. Adam, do you feel that your server could handle a big rally?

This would allow us to use IronBit's server as a second server (in addition to Carlos's server) on port 300 for 400<k<=1001 for n>370K and for him to now reserve however much he wants of that range.

The question that I'll ask everyone is: Whose server is the best to use on 300<k<400 for n>333.3K for the next rally? That will dictate how we distribute current reservations.



Gary

Since IronBits's server is a big powerful quad-core, running Ubuntu so it shouldn't have any problems with connection limits, I think we should put the 300<k<400, n>333.3K range on a server hosted by him, since his machine should be able to handle a rally best. We can then set up AES's server as a second server (in addition to Carlos's port 300 server) for the 400<k<1001, n>370K range.

Then, meanwhile, Gary's server can do 300<k<400, n<333.3K.

When higher (n>600K) ranges become available, we can then decide which server should be able to handle such a server in addition to its current load at the time, and then put those higher ranges on that server.

What does everyone think?

Anon

[gd_barnes]

Originally Posted by Anonymous

Since IronBits's server is a big powerful quad-core, running Ubuntu so it shouldn't have any problems with connection limits, I think we should put the 300<k<400, n>333.3K range on a server hosted by him, since his machine should be able to handle a rally best. We can then set up AES's server as a second server (in addition to Carlos's port 300 server) for the 400<k<1001, n>370K range.

Then, meanwhile, Gary's server can do 300<k<400, n<333.3K.

When higher (n>600K) ranges become available, we can then decide which server should be able to handle such a server in addition to its current load at the time, and then put those higher ranges on that server.

What does everyone think?

Anon

I like this idea.

OK, Iron Bits, how much data do you think you need for drive 1 to keep you busy for 3 days? n=200 is about 2250-2300 k/n pairs.

[IronBits]

Originally Posted by gd_barnes

I like this idea.

OK, Iron Bits, how much data do you think you need for drive 1 to keep you busy for 3 days? n=200 is about 2250-2300 k/n pairs.

I have no clue, depends on how many folks are going to come get work right?
Beyond is hitting port 500 to help clean up the 'mess' ;)
I have my two quads on it as well.

I'm here when you need/want me. I'm not going to pretend I know anything about the project, however, I can run you a server on 3 different ports if you like and send you the results and let you know if there is found prime. :)

[gd_barnes]

Originally Posted by IronBits

I have no clue, depends on how many folks are going to come get work right?
Beyond is hitting port 500 to help clean up the 'mess' ;)
I have my two quads on it as well.

I'm here when you need/want me. I'm not going to pretend I know anything about the project, however, I can run you a server on 3 different ports if you like and send you the results and let you know if there is found prime. :)

IronBits,

You're right. We should be figuring that out!

I just sent a PM to Carlos and Anon asking their opinion about it. What I told them is that I think it's going to be too confusing for people to move servers on and off of ranges. Here's what I'm suggesting to them: We won't need the additional servers for a few days. I'll be done sieving k=300-400 for n=260K-600K this weekend. Shortly after that, I think I'd like to put n>333.3K on one of your remaining servers in preparation for a rally the weekend of the 8th. It sounds like your servers could easily handle a rally.

I'd also like to put your other server on k=300-400 for n=260K-333.3K, the lower range, at the same time as the higher range.

In the future:

I'd like to get k=300-400 caught up to k=400-1001 and then combine the ranges. It'll make it easier to manage. At that point, sieving on n=600K-1M should be ready for k=300-400 and perhaps we can just have 2 servers on n=333.3K-600K and one on n>600K or visa versa. Note that only a few k's will be ready to search for n>600K; perhaps 10 of them. At that point, we'll only be searching the k's that have been previously searched to n=600K.


Gary

[gd_barnes]

Originally Posted by AES

I believe nplb.rieselprime.org port 300 can now safely serve any work deemed appropriate.

We can also test feeding an llrserver remotely from the database at nplb.rieselprime.org if anyone is interested. It worked on my test over a satellite internet link with 1000+ ms latency. However, there were only 5 cores demanding work.

Originally Posted by Anonymous

Since IronBits's server is a big powerful quad-core, running Ubuntu so it shouldn't have any problems with connection limits, I think we should put the 300<k<400, n>333.3K range on a server hosted by him, since his machine should be able to handle a rally best. We can then set up AES's server as a second server (in addition to Carlos's port 300 server) for the 400<k<1001, n>370K range.
Anon

Adam,

I will reserve an n=5K range for your LLRnet server on drive 1. Look for the file in the first post of that thread. When you have it, let me know and I'll remove the link.

Edit: n=395K-400K has been reserved for the server.


Gary

[BlisteringSheep]

Are there still more servers needed? I've got two Linux boxes and one Windows box that could host one server each.

If they aren't needed (sounds like IB's probably got it covered, and his server host is stronger than mine would be), would it be better for the next rally (or whenever I'm running the octo) if I set up a personal proxy to minimize the frequency of connects? I should probably do that anyway, because right now the clients always pause between tests to send their result & fetch a new one (even with the cache - they always keep that many ready). Suggestions?

[IronBits]

You should run a personal proxy, if for nothing else, at least that octo. ;)
Gives you a lot more control and helps to manage your bandwidth.

[gd_barnes]

Originally Posted by BlisteringSheep

Are there still more servers needed? I've got two Linux boxes and one Windows box that could host one server each.

If they aren't needed (sounds like IB's probably got it covered, and his server host is stronger than mine would be), would it be better for the next rally (or whenever I'm running the octo) if I set up a personal proxy to minimize the frequency of connects? I should probably do that anyway, because right now the clients always pause between tests to send their result & fetch a new one (even with the cache - they always keep that many ready). Suggestions?

Not at this time. I kind of 'over suggested' things as far as servers are concerned. See a new topic started by Carlos. But thanks for offering!!

I'll leave that to the server gurus to answer your question about setting up a personal proxy.

[tnerual]

Originally Posted by BlisteringSheep

Are there still more servers needed? I've got two Linux boxes and one Windows box that could host one server each.

If they aren't needed (sounds like IB's probably got it covered, and his server host is stronger than mine would be), would it be better for the next rally (or whenever I'm running the octo) if I set up a personal proxy to minimize the frequency of connects? I should probably do that anyway, because right now the clients always pause between tests to send their result & fetch a new one (even with the cache - they always keep that many ready). Suggestions?

the good point of a personnal proxy is the way you can change of ranges (connecting to another master server)

using normal connection, you have to go on every computer, changing the port.
using a proxy, you only change the masterserver port on the proxy, and all client stay exactly as they are ... no change at all

your proxy will also be usefull to manage the masterserver ping delay (no delay to contact your proxy server locally.

[IronBits]

Would it be possible to get some work for port 5000 so I can test the CentOS 4.5 64-bit (Q6600 w/8GB of ram) Server setup and see how that's going to work out please?
top-5000 range

On my port 500 Windows XP SP2 'Server' where we are cleaning up 'Gary's mess' ;)
If I use maxConnections = 60 I don't seem to be having any problems.
Virtual memory = 120,000
Peak memory = 57,000
Peak Page file = 55,000