LLRnet server rally 400<k<1001 June 20-22

[mdettweiler]

Hi all,

Everyone ready for another rally? Well, you'd better be, because we're doing one on the 400<k<1001 (1st Drive) range this weekend.

As discussed in the "thoughts about upcoming LLRnet rallies" thread, the rally will run for 48 hours, from 7 PM GMT on June 20th to 7 PM GMT June 22 (i.e. 2 PM CDT or 3 PM EDT). It will primary run on servers configured for the 400<k<1001 range, though results from port 5000 (configured for the 300<k<400 range) will be counted, too. (A little word of warning, though, if you plan to run on the 300<k<400 range during the rally: the n-ranges that we are currently testing at for 300<k<400 are somewhat higher than the respective ranges for 400<k<1001, and thus each individual number takes noticeable longer to run. Thus, if you're concerned primarily with your score in the rally standings that we will publish after the rally, then your best bet is to stick to 400<k<1001, where you can get more k/n pairs done in a given amount of CPU time.)

The server info for all our 400<k<1001 range servers is as follows:
-------------------------------------------
server = "nplb.rieselprime.org"
port = 300
-------------
server = "llrnet.ironbits.net"
port = 400
-------------
server = "llrnet.ironbits.net"
port = 8500
-------------
server = "nplb.dynip.telepac.pt"
port = 400
-------------
server = "samband.mine.nu"
port = 6
-------------------------------------------

Edit: This rally is now officially over. Thanks all for a great rally!

General info. on setting up a connection to an LLRnet server can be found here. The above info. should be put in the llr-clientconfig.txt file.

I'd like to request that everyone participating in the rally say here how many cores they plan to bring to the rally, preferably as early as possible. (Yours truly will bring two high-speed cores, as usual. ) Hopefully, with proper planning as to how many servers we need and how much work to load into them, we can avoid another fiasco like the one we had last rally. If you're bringing a significant number of cores (~10+), please be especially sure to say so, and make sure to specify which range (400<k<1001 or 300<k<400) you'd prefer--we can then tell you which server your machines would fit best on during the rally. Hopefully by managing the "heavy hitter" resources like this, we can keep the servers reasonably balanced load-wise, and "little guys" with <5 cores (like me!) can just slip in wherever they feel like.

As for goals for this rally, that is made a bit more complicated by the fact that our individual servers within the 400<k<1001 range are somewhat spread out in terms of n-ranges--so, how about we set a goal of finishing all the ranges currently loaded into the servers (458K-463K for port 400, 473K-478K for port 8500, and 478K-485K for port 300), thus closing the "gaps" in testing ranges, and getting the servers all moved on to new ranges that will hopefully be closer to each other.

So, let's bring on that firepower!

Anon

P.S.: Conducive to this goal of closing n-range gaps between servers, I think it would be wise to refill all the servers (except, of course, port 8500, which we were planning to run dry) at once, so that the ranges assigned to all the servers will be budged up against each other. Carlos, while I am getting work to refill all of the active servers, I'll also send you a file for the nplb.dynip.telepac.pt port 400 server that you've offered to host.

[kar_bon]

i will throw in all cores of my quad, perhaps one other smaller core. currently running on IB400 server, can change to A300 as well.

and another hint:
to avoid many 'open' pairs after the rally, please everyone be sure to clear the workfile of all clients, so we have not to wait some days for a second submission of these pairs. thanks.

[mdettweiler]

Here's a table representing all the cores that users plan to bring to the rally, which I will try to keep updated as more people provide this information:

User           Cores              Range preference   Server
-----------------------------------------------------------
Anonymous      2 fast             400<k<1001         1 core on C400, 1 core on AES300
kar_bon        4 fast, 1 slower   400<k<1001         IB400 (or AES300)
Flatlander     4 fast             400<k<1001         AES300
Mini-Geek      2 slower           400<k<1001         IB400
gd_barnes      3 fast, 4 slow     400<k<1001         AES300
gd_barnes      5 fast             400<k<1001         IB400
glennpat       10 fast            400<k<1001         3 on IB5000, 3 on C400, 2 on IB400, 2 on AES300
Lennart        66+ fast           400<k<1001         50+ on L6, 8 on AES300, 8 on IB400
em99010pepe    4 fast             300<k<400          IB5000
HelixVonSmelix 2 or 3 unknown     300<k<400          IB5000
Beyond         6 unknown          400<k<1001         C400
----------------------
Total on IB400:  22
Total on AES300: 22
Total on C400:   10
Total on IB8500: 0
Total on L6: 50+
------
Total on IB5000: 9 or 10

[Flatlander]

I'll bring 4 fast cores to 400<k<1001.

[gd_barnes]

I was going to suggest delaying the rally 1-2 weeks. I'm out of town again this coming weekend but will be back in town the following 2 weekends.

If it's this weekend, I can only bring 5 cores. I'm still not going to have time to set up the remote desktop before I leave and I have too many other reservations to stop them for 8 days and put all my cores on the rally servers for that long of a period.

What hurt me was the power outage that I had 3 hours after I left last time for 8 days. My 28 cores were idled for almost the entire time leaving all of my reservations well behind where they should be.


Gary

[mdettweiler]

Originally Posted by gd_barnes

I was going to suggest delaying the rally 1-2 weeks. I'm out of town again this coming weekend but will be back in town the following 2 weekends.

If it's this weekend, I can only bring 5 cores. I'm still not going to have time to set up the remote desktop before I leave and I have too many other reservations to stop them for 8 days and put all my cores on the rally servers for that long of a period.

What hurt me was the power outage that I had 3 hours after I left last time for 8 days. My 28 cores were idled for almost the entire time leaving all of my reservations well behind where they should be.


Gary

Unfortunately, the next two weekends won't work for me at all. I'll be taking a weeklong vacation from the ~28th to the ~5th (not sure exactly which days I'll be leaving and returning), and may or may not have any access to mersenneforum at all. If I do have any access, it will be quite limited, and I won't have any access to my results processing tools whatsoever. Most likely I'll only have one or two older laptops with me on vacation, and they'll only be able to connect through dial-up, so no remote controlling my home machine to use my processing tools.

By contrast, Gary, you have a moderate degree of mersenneforum access on your trips, so it should theoretically have a lesser impact on the rally if you weren't home than if I wasn't home.

So, how about we have the rally this weekend, and then hold off on our next one until I am back--that way, we'll have the maximum possible amount of admin personnel on hand during the rally. Gary, considering the situation with your various reservations, don't fret about how many cores you put on the rally--just put on what you can spare, don't worry about have reservations make way for the rally.

[Mini-Geek]

I'll bring my two Athlon X2 2.5 GHz cores to it, whatever time it is (as always). I guess you'd consider that fast (at n=485K, 32K FFT, I take ~455 seconds per pair). I'll run it on 400<k<1001.
Maybe in that chart we should also keep track of what server your cores will be running on, to try to balance it as much as possible. If you decide to do that, put me down for 2 fast cores on IB400. Oh, and that chart should be moved to the first post.

[em99010pepe]

Anon,

My server is up and running.

Mini-Geek,

That's slow, I get 224 seconds with LLRnet on n=511k (drive 3).
Edit: Tested the server and I get 226 seconds on n=505k 400<k<1001.

[mdettweiler]

Originally Posted by em99010pepe

Anon,

My server is up and running.

Mini-Geek,

That's slow, I get 224 seconds with LLRnet on n=511k (drive 3).
Edit: Tested the server and I get 226 seconds on n=505k 400<k<1001.

Great! I've already added it to the "LLRnet Servers for NPLB" thread; I'll add it to the first post of this thread as soon as I submit this post.

[mdettweiler]

Originally Posted by Mini-Geek

I'll bring my two Athlon X2 2.5 GHz cores to it, whatever time it is (as always). I guess you'd consider that fast (at n=485K, 32K FFT, I take ~455 seconds per pair). I'll run it on 400<k<1001.
Maybe in that chart we should also keep track of what server your cores will be running on, to try to balance it as much as possible. If you decide to do that, put me down for 2 fast cores on IB400. Oh, and that chart should be moved to the first post.

Okay, yeah, that sounds like a good idea. I'll add that field to the table shortly. (For those who don't really care which server within a given range you go on, just pick whichever one has the least amount of people on it already.)

If you have more than one core in the rally, though, you may want to considering balancing out your cores over all the servers in your chosen range. I'm planning to put one of my cores on one server and the other on another (I'll play it by ear as to which server I go on, based on which ones have the lowest load at the time I switch my cores over to rally configuration). This can also provide some failover protection, so that if one server goes down at least part of your machine is idle.

If anyone else plans to wait and see which server has the lowest load, and go on that (rather than picking out a server ahead of time) just say so and I'll mark you with a * in the server field in the table.

[gd_barnes]

Originally Posted by Anonymous

Unfortunately, the next two weekends won't work for me at all. I'll be taking a weeklong vacation from the ~28th to the ~5th (not sure exactly which days I'll be leaving and returning), and may or may not have any access to mersenneforum at all. If I do have any access, it will be quite limited, and I won't have any access to my results processing tools whatsoever. Most likely I'll only have one or two older laptops with me on vacation, and they'll only be able to connect through dial-up, so no remote controlling my home machine to use my processing tools.

By contrast, Gary, you have a moderate degree of mersenneforum access on your trips, so it should theoretically have a lesser impact on the rally if you weren't home than if I wasn't home.

So, how about we have the rally this weekend, and then hold off on our next one until I am back--that way, we'll have the maximum possible amount of admin personnel on hand during the rally. Gary, considering the situation with your various reservations, don't fret about how many cores you put on the rally--just put on what you can spare, don't worry about have reservations make way for the rally.

That's fine. I'll put my 5 cores on port 300 (currently on port 400) shortly before I leave in the morning. It has the most remaining work at the moment and I'm sure you'll keep it loaded up if it runs low.

Is there a way we can have Beyond in the rally? In other words, is there a way to count how much he processes of his 486K-496K range during the rally hours?


Gary