mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2021-04-03, 16:13   #1
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

1110100000102 Posts
Default Registered Memory (-- edit: --) is Not the Problem

Edit: It turned out to be an issue with the current revision of CADO-NFS, rather than the first time use of Registered memory.

I recently acquired a dual Xeon machine and "tried" to put it to work with my others. It came with Registered ECC RAM. In trying to run the CADO-NFS client, the 16GB is overwhelmed. On other machines, including two Xeon dual core, the memory use is only ~4GB per client.

Could this be a result of the RAM being Registered?

In my research, I have seen references to Registered RAM running slower in some applications, but is it possible that it is causing such an increase in need?

Last fiddled with by EdH on 2021-04-04 at 15:05 Reason: Clarify real problem.
EdH is offline   Reply With Quote
Old 2021-04-04, 00:01   #2
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

27·5 Posts
Default

Quote:
Originally Posted by EdH View Post
I recently acquired a dual Xeon machine and "tried" to put it to work with my others. It came with Registered ECC RAM. In trying to run the CADO-NFS client, the 16GB is overwhelmed. On other machines, including two Xeon dual core, the memory use is only ~4GB per client.

Could this be a result of the RAM being Registered?

In my research, I have seen references to Registered RAM running slower in some applications, but is it possible that it is causing such an increase in need?
Probably not, but what do you mean by 'overwhelmed' ?
tServo is offline   Reply With Quote
Old 2021-04-04, 02:24   #3
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2·3·619 Posts
Default

Quote:
Originally Posted by tServo View Post
Probably not, but what do you mean by 'overwhelmed' ?
A process that takes about 4GB on all my other machines used up over 14GB of Registered memory on the "new" (to me) machine and went into swap memory. The machine was just short of seeming to be locked up. I had to issue commands via ssh and wait several minutes for them to be accepted. I removed one of the CPUs and added the two memory modules from that assembly to the single one and tried again and it acted the same.

The machine is a Z620 with Xeon e5-2630 CPU(s) AND 16GB of Registered 4x4GB modules. I tried to use some non-ECC modules as a test, but it won't work with non-ECC. I have no available ECC ATM to test with, but I do have some on the way.

Last fiddled with by EdH on 2021-04-04 at 02:25 Reason: just 'cause
EdH is offline   Reply With Quote
Old 2021-04-04, 04:45   #4
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

34×59 Posts
Default

Is it possible you forgot to change threads-per-client, and it spawned like 12 clients rather than 4?
VBCurtis is offline   Reply With Quote
Old 2021-04-04, 12:07   #5
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

E8216 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Is it possible you forgot to change threads-per-client, and it spawned like 12 clients rather than 4?
No. it's just one client, shown as "las" in "top." It's running the same script all the other machines are running and all the software is the same, other than currency. Do you think it may be the latest revision of CADO-NFS is causing it? I really think that would be stretching things. But this version is whatever it's up to as of a couple days ago.
EdH is offline   Reply With Quote
Old 2021-04-04, 12:28   #6
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

27×5 Posts
Default

If this is the first time you have tried your script with a different dual processor machine, I would say that is more likely the cause than the type of memory.
The link below has a nice description of how buffered memory works:
https://www.servethehome.com/unbuffe...udimms-rdimms/

Also, vendors such as HP sometimes run lots of junky tasks. Are any of those consuming too much memory?

Last fiddled with by tServo on 2021-04-04 at 12:33
tServo is offline   Reply With Quote
Old 2021-04-04, 13:19   #7
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2·3·619 Posts
Default

Quote:
Originally Posted by tServo View Post
If this is the first time you have tried your script with a different dual processor machine, I would say that is more likely the cause than the type of memory.
The link below has a nice description of how buffered memory works:
https://www.servethehome.com/unbuffe...udimms-rdimms/

Also, vendors such as HP sometimes run lots of junky tasks. Are any of those consuming too much memory?
I'd already found the linked page in my researh, but thank you for posting it. There shouldn't be any HP junk on this machine that isn't on the other nearly identical one.

I tried running the client with 2 threads as opposed to saturation, as the others are running (4, 8, 24) and this machine is changing its behaviour based on the threads. Two threads starts out at the ~4GB that all the others are using, but moves to ~8GB before it completes the run. A run with 4 threads, nearly immediately moves to ~10GB. I'm going to try 6 threads next and then try a regression on my CADO-NFS install and see what happens.

Thanks for all help!
EdH is offline   Reply With Quote
Old 2021-04-04, 14:14   #8
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2×3×619 Posts
Default

Progress (or, perhaps regression)!

I updated one of my other machines to the most recemt CADO-NFS and it ran the memory up and locked up so I couldn't communicate with it in a timely fashion. Since it is headless and remote, I am now waiting fo a "pkill" command to free it, which may take a while.

At this point, I guess the CADO-NFS package is the "culprit." I'll try to verify it by changing the revision in the new machine.
EdH is offline   Reply With Quote
Old 2021-04-04, 14:40   #9
PhilF
 
PhilF's Avatar
 
Feb 2005
Colorado

617 Posts
Default

Quote:
Originally Posted by EdH View Post
Progress (or, perhaps regression)!

I updated one of my other machines to the most recemt CADO-NFS and it ran the memory up and locked up so I couldn't communicate with it in a timely fashion. Since it is headless and remote, I am now waiting fo a "pkill" command to free it, which may take a while.

At this point, I guess the CADO-NFS package is the "culprit." I'll try to verify it by changing the revision in the new machine.
That makes a lot more sense to me.

Otherwise, if it were to be confirmed, you would have been the first case of RAM type changing the software's memory requirements that I've ever come across in over years.

Last fiddled with by PhilF on 2021-04-04 at 15:23 Reason: Feeble attempt at clarity
PhilF is online now   Reply With Quote
Old 2021-04-04, 15:01   #10
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

72028 Posts
Default

Quote:
Originally Posted by PhilF View Post
That makes a lot more sense to me.

If confirmed, you would have been the first case of RAM type changing the software's memory requirements that I've ever come across in over years.
And, I consider it confirmed!

I moved the new machine back to the same revision as the others and it is running along in the manner I originally expected.

Now the dilemma of whether to report my adventure. . .

Thanks to all for help in this. I should have considered the possibility of it being CADO-NFS in the first place, but Registered memory is a new area to me and "I thought" CADO-NFS was familiar.
EdH is offline   Reply With Quote
Old 2021-04-04, 20:50   #11
charybdis
 
charybdis's Avatar
 
Apr 2020

263 Posts
Default

Quote:
Originally Posted by EdH View Post
Now the dilemma of whether to report my adventure. . .
Absolutely. This sounds like a significant bug and the sooner the CADO developers find out about it, the better.
charybdis is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Registered Assignments storm5510 PrimeNet 8 2017-07-12 03:52
Registered for the smallest exponents Chuck PrimeNet 5 2014-02-28 02:23
Problem: 0 registered computers PodolskyV Information & Answers 2 2012-01-03 20:56
Only one instance of mprime is registered Unregistered Information & Answers 8 2008-04-09 00:24
Log-out for registered users Budgie Forum Feedback 19 2007-03-01 13:12

All times are UTC. The time now is 20:25.

Wed May 12 20:25:24 UTC 2021 up 34 days, 15:06, 1 user, load averages: 2.19, 2.28, 2.16

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.