mersenneforum.org  

Go Back   mersenneforum.org > Other Stuff > Open Projects > y-cruncher

Reply
 
Thread Tools
Old 2019-10-22, 17:46   #12
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

8016 Posts
Default

Yes, I guess Ctrl+D should be minimal priority for now.

By the way, I had a computation running on Linux with 9501 and it simply exited in the middle of the computation. I guess, but cannot assure that RAM was not the issue. If there is no error message, like "killed by...", do you have any idea what might have caused that? I went on running golden ration on 5 trillion digits, but before restarting, I want to see what might went wrong. Since there are -- understandably -- no checkpoints on square roots, I guess I cannot do a lot if the issue is not one of y-cruncher (since I assume it's not)?
kruoli is offline   Reply With Quote
Old 2019-10-23, 06:15   #13
Mysticial
 
Mysticial's Avatar
 
Sep 2016

32910 Posts
Default

Quote:
Originally Posted by kruoli View Post
Yes, I guess Ctrl+D should be minimal priority for now.

By the way, I had a computation running on Linux with 9501 and it simply exited in the middle of the computation. I guess, but cannot assure that RAM was not the issue. If there is no error message, like "killed by...", do you have any idea what might have caused that? I went on running golden ration on 5 trillion digits, but before restarting, I want to see what might went wrong. Since there are -- understandably -- no checkpoints on square roots, I guess I cannot do a lot if the issue is not one of y-cruncher (since I assume it's not)?
That's strange. I've also never seen things die on Linux without some sort of "killed" or "segfault" message. When it dies from OOM, it will (at least on out-of-the-box Ubuntu) will say killed.

Whenever the program exits abnormally due to internal causes, it will always pause and wait for user-input unless it's suppressed via the "pause:-2" flag.

Is it possible that the process got tossed into the background and thus no output at all?

Last fiddled with by Mysticial on 2019-10-23 at 06:15
Mysticial is offline   Reply With Quote
Old 2019-10-24, 12:12   #14
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

27 Posts
Default

You mean by something like Ctrl+Z, bg, fg, nohup? No, that was not the case. It ran inside of a tmux environment which I always use for having long running tasks. It will keep everything written out (up to a backwards limit), so it should have kept such messages (and has it on other systems before). I guess, it might have to do with that weird custom Linux that box is running on and there is little I can do about that (the same with the missing NUMA).

Since you never experienced that, I'll guess we set it on hold until we can reproduce it on a "normal" Linux machine? I'm going forward and start the same computation somewhere else soon.
kruoli is offline   Reply With Quote
Old 2019-10-29, 12:53   #15
DualSocketWorld
 
Oct 2019

110 Posts
Default About swap I/O

I hope you to make it faster swap I/O programing (now it use only one thread), in case of using multiple RAID cards and/or multiple NVMe, Optane or someting like that.
DualSocketWorld is offline   Reply With Quote
Old 2019-10-29, 16:33   #16
Mysticial
 
Mysticial's Avatar
 
Sep 2016

7×47 Posts
Default

Quote:
Originally Posted by DualSocketWorld View Post
I hope you to make it faster swap I/O programing (now it use only one thread), in case of using multiple RAID cards and/or multiple NVMe, Optane or someting like that.
Not sure if I understand what you're referring to.

Right now, swap mode uses one thread per logical path. So you can get multiple threads by using multiple logical paths - thus it's advantageous to split up your arrays and let y-cruncher handle each drive manually.

The reason why it only uses one thread per logical path is for 2 reasons:
  1. The program right now assumes that parallel access to a single drive is neither possible nor beneficial (as is the case for hard drives). Thus it was never designed to parallelize access to a single path.
  2. The underlying API calls to perform disk I/O are inherently single-threaded. There's no multi-threaded version of ReadFile/WriteFile (Windows) and read()/write() (Linux/Posix)* that I'm aware of.

*At least in Windows (with the right API flags), it is possible to issue parallel ReadFile/WriteFile calls to the same file in different threads. But the situation for Linux is more unclear. Either way, it's an area of research that's not on my immediate to-do list.

Last fiddled with by Mysticial on 2019-10-29 at 16:38
Mysticial is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
LLR beta Version 3.8.13 (deprecated) Jean Penné Software 111 2015-01-26 21:41
Prime95 beta version 28.4 Prime95 Software 20 2014-03-02 02:51
Prime95 beta version 28.3 Prime95 Software 68 2014-02-23 05:42
Beta version 24.12 available Prime95 Software 33 2005-06-14 13:19
Beta version of PRP Prime95 PSearch 15 2004-09-17 19:21

All times are UTC. The time now is 22:41.

Thu May 28 22:41:10 UTC 2020 up 64 days, 20:14, 1 user, load averages: 1.05, 1.33, 1.43

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.