On Mon, 2018-02-05 at 10:15 +1300, Ben Caradoc-Davies wrote:
> On 05/02/18 09:49, Stephen P. Molnar wrote:
> > They seem to happen when I am
> > running four or more apps at the same time.
>
> I would never expect to see orphaned inodes except after a system
> crash
> or kernel memory corruption. How did you test your CPU and RAM? Do
> you
> see any other symptoms such as segfaults that could suggest memory
> problems under concurrent load? How long have you seen this problem?
> I
> see you are using ext4; are the inodes on these filesystems? ext4 is
> very well tested and robust.
>
> My preferred memory test for my 4-core (8-thread) Kaby Lake i7 is to
> run
> concurrent "memtester" instances equal to the number of cores (4 in
> my
> case), concurrent with "stress" equal to the number of cores ("stress
> -c
> 4" in my case). This workout detected memory problems not found by
> other
> tools such as "memtest86+" or "mprime -t".
>
> Other hardware issues to consider are overheating (addressed with
> better
> cooling and thermald) and power supply problems which may only be
> evident at load. Is your system prime stable (i.e. runs with "mprime
> -t"
> (with AVX disabled) for many hours)? How do you monitor system
> temperature?
>
> Kind regards,
>
Thanks for your reply.
As a matter of fact I did get an overheating warning from the OS when
running a rather large organic molecule on the Orca package with 8
threads. I upgraded the CPU cooler to a Hyper 212 EVO and the problem
went away.
--
Stephen P. Molnar, Ph.D.
Consultant
www.molecular-modeling.net
(614)312-7528 (c)
Skype: smolnar1
