mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > Cunningham Tables

Reply
 
Thread Tools
Old 2021-10-07, 22:12   #100
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2·29·71 Posts
Default

Quote:
Originally Posted by kruoli View Post
. . .
Unfortunately, I do not have a static IP address, but I have a domain with dynamically updated DNS records of my IP. Inside my home network, I usually use host names instead of IP addresses, so I assume CADO will be fine using a domain, but I'm not sure how it would handle a change of IP while a work unit is processed. The IP gets renewed daily, the DNS record is updated immediately afterward.

I already have some services accessible through my router via port forwarding, so I assume this should not be a problem for us.
I'll get corrected if I'm wrong, here, but I think you'll need clients to provide their external IPs so you can whitelist them for your CADO-NFS server.
EdH is offline   Reply With Quote
Old 2021-10-07, 22:20   #101
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

28·3 Posts
Default

IIRC, there is also some mode where it accepts all IPs, but I guess that's insecure and thus not recommended.

But whitelisting client IPs should not be a problem even if my own IP is not static, correct? Here, we should have a look at how other team sieves handled that. I should be able to adopt that specific aspect.
kruoli is online now   Reply With Quote
Old 2021-10-07, 22:26   #102
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

10000000101102 Posts
Default

Quote:
Originally Posted by kruoli View Post
IIRC, there is also some mode where it accepts all IPs, but I guess that's insecure and thus not recommended.

But whitelisting client IPs should not be a problem even if my own IP is not static, correct? Here, we should have a look at how other team sieves handled that. I should be able to adopt that specific aspect.
VBCurtis is the only one I've run clients with. I believe I needed an IP and port with him and to supply my external IP. Accepting all IPs probably isn't recommended, but it is probably available. You will also have to provide an SSL certificate or run with no SSL.

I guess VBCurtis will probably chime in with any corrections or additions.
EdH is offline   Reply With Quote
Old 2021-10-08, 00:40   #103
Plutie
 
"Evan"
Dec 2020
Montreal

23·32 Posts
Default

Oh yeah, forgot that you had wanted this one. I can still pledge a bunch of CPU-time nonetheless, so keep me posted!
Plutie is offline   Reply With Quote
Old 2021-10-08, 01:55   #104
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

507310 Posts
Default

On the C207 Cunningham team sieve, I chose to accept all IP addresses; I had originally planned to whitelist by IP (with :16, so that the 3rd/4th numbers that often vary from service providers on client restarts would not matter). Then quite a few more people joined, and I was out of the country, so I set it to accept any IP.

That's part of why I also kept the server name/address on PM only- I figured there was only a very small chance that someone would both find the open port by sniffing *and* send a packet that CADO would recognise/reply to. I had no problems during the 60+ days the CADO server was up with random accesses, though the CADO server itself needed a reboot fairly often because the database wasn't great for such a big project. One of the CADO folks suggested I change databases for a team sieve, but such meddling is above my measly pay grade (having never knowingly admin'ed a database previously).

As for server side, I think a dynamic IP with a domain name should work- clients should work fine pointed at e.g. my.domain.com:portnumber

The port number needs to be fixed when you fire up the CADO server.
VBCurtis is offline   Reply With Quote
Old 2021-10-09, 20:04   #105
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

10011110100012 Posts
Default

Here are the params I have for this job:
Code:
tasks.lim0 = 500000000
tasks.lim1 = 300000000
tasks.lpb0 = 33
tasks.lpb1 = 34
tasks.sieve.mfb0 = 66
tasks.sieve.mfb1 = 98
tasks.sieve.ncurves0 = 25
tasks.sieve.ncurves1 = 20
tasks.A = 30
tasks.sieve.las.threads = 4 
tasks.sieve.adjust_strategy = 2
tasks.qmin = 80000000
tasks.sieve.rels_wanted = 1300000000
tasks.sieve.qrange = 2000
Client memory use is around 5.5GB. I tested smaller lim's, but memory use barely changes and sec/rel barely changes, but yield is quite a bit better with lim's this large. I forecast Q from 80-600M to reach 1300M raw relations, which should be enough to get a matrix under 60M size.
The q-range of 2000 completed a 4-threaded workunit on my Ivy Bridge 2.6Ghz in around 30 minutes; imo this is the smallest workunit that makes sense.

EDIT: I added "tasks.sieve.adjust_strategy = 2" above. I forgot this useful setting on my test-sieving too, but I trust that it would not change the choice of fastest parameters.

Last fiddled with by VBCurtis on 2021-10-09 at 21:05
VBCurtis is offline   Reply With Quote
Old 2021-10-19, 11:56   #106
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

28·3 Posts
Plus Ready to go!

The team sieve server setup is ready. Anyone who wants to contribute can PM me for the details.

What you should know in advance:
  • We use CADO's git commit 9e81c910 (e.g. perform a fresh clone/checkout/pull/etc., then git reset --hard 9e81c910). Please make sure you can build this on your machine and that make check comes up without errors.
  • Each client uses four threads and around 7 GB of RAM. VBCurtis reported 5.5 GB, but I guess this was using the default tasks.sieve.adjust_strategy.
  • Work units will time out after ten hours. A single work unit (running under WSL2) took my AMD 3800X 12 minutes, so ten hours should not be a problem for any Intel/AMD CPU with at least four cores that can handle 8 GB and more of main memory; even old Atoms should manage this. Problems could occur if higher priority processes use up lots of CPU cycles.

Last fiddled with by kruoli on 2021-10-19 at 12:39 Reason: Title added. Missing bracket added.
kruoli is online now   Reply With Quote
Old 2021-10-19, 12:44   #107
charybdis
 
charybdis's Avatar
 
Apr 2020

547 Posts
Default

Thanks Oliver for getting everything set up!

Using commit 9e81c910 isn't essential, but I would highly recommend that everyone upgrades to it, because it speeds up sieving by several percent (possibly more on AVX-512 systems).

Quote:
Originally Posted by kruoli View Post
Each client uses four threads
As usual, if you want to run a client with a different number of threads, you can use the --override flag: for example, if you want to run a 3-threaded client, add --override t 3 to your client invocation.

Last fiddled with by charybdis on 2021-10-19 at 12:45
charybdis is offline   Reply With Quote
Old 2021-10-19, 12:54   #108
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

28·3 Posts
Default

Thanks to you and VBCurtis for supporting my setup process and testing!

There is still a lot to learn for me, e.g. what can be changed client-side etc.
kruoli is online now   Reply With Quote
Old 2021-10-21, 00:15   #109
charybdis
 
charybdis's Avatar
 
Apr 2020

547 Posts
Default

I've spotted an oddity: adjust_strategy=2 appears to be clearly worse until just above Q=120M. I haven't seen this happen before; perhaps it's a consequence of the high skew? In any case, I'd suggest changing this in the params file for now until we reach 120M and restarting the server. As long as the port stays the same, the clients should reconnect; I don't think anything else needs to be done (@VBCurtis?)
charybdis is offline   Reply With Quote
Old 2021-10-21, 02:12   #110
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2·29·71 Posts
Default

Quote:
Originally Posted by charybdis View Post
I've spotted an oddity: adjust_strategy=2 appears to be clearly worse until just above Q=120M. I haven't seen this happen before; perhaps it's a consequence of the high skew? In any case, I'd suggest changing this in the params file for now until we reach 120M and restarting the server. As long as the port stays the same, the clients should reconnect; I don't think anything else needs to be done (@VBCurtis?)
I would change the setting in the latest snapshot and restart using the restart command line in the log with the edited snapshot. All else should take care of itself. Restarting from scratch might work also, but I always use the suggested restart.

Last fiddled with by EdH on 2021-10-21 at 02:13
EdH is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Using 16e on smaller numbers fivemack Factoring 3 2017-09-19 08:52
NFS on smaller numbers? skan YAFU 6 2013-02-26 13:57
Bernoulli(200) c204 akruppa Factoring 114 2012-08-20 14:01
checking smaller number fortega Data 2 2005-06-16 22:48
Factoring Smaller Numbers marc Factoring 6 2004-10-09 14:17

All times are UTC. The time now is 09:13.


Tue Dec 7 09:13:41 UTC 2021 up 137 days, 3:42, 0 users, load averages: 1.98, 1.58, 1.50

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.