[Wien] Fragmented RAM slowdown of lapw1_mpi
Laurence Marks
laurence.marks at gmail.com
Sat Dec 21 14:08:25 CET 2019
As an addendum, this appears to be a common problem, e.g.
https://library.softwareverify.com/memory-fragmentation-your-worst-nightmare/
_____
Professor Laurence Marks
"Research is to see what everybody else has seen, and to think what nobody
else has thought", Albert Szent-Gyorgi
www.numis.northwestern.edu
On Fri, Dec 20, 2019, 15:20 Laurence Marks <laurence.marks at gmail.com> wrote:
> I believe that I've finally managed to trace a generic slowdown of mainly
> lapw1_mpi. It's been around since 2013 and is probably related to how files
> are used in WIEN2k & NFS. Maybe someone knows a friendly OS person who can
> provide some information.
>
> The symptom is that lapw1 -p with mpi over a period of some days slows
> down, with the Wall time increasing although the CPU time also increases a
> little. It appears that this is related to fragmentation of the RAM, so
> large matrices are starting to occupy disjointed locations in memory and
> are therefore slower. A cure appears to be to run the builtin RAM
> defragger, "echo 1 >/proc/sys/vm/compact_memory" [1]. This probably does
> not need to be done more than every few days or perhaps only once a week.
>
> Anyone come across anything similar? It may well have to do with how NFS &
> Read/Write I/O is done.
>
> [1] WARNING: do not attempt this unless you are quite linux experienced.
> An alternative to check is to do "cat /proc/buddyinfo", looking up what
> that information means (e.g.
> https://www.uninformativ.de/blog/postings/2017-12-23/0/POSTING-en.html ).
>
> --
> Professor Laurence Marks
> Department of Materials Science and Engineering
> Northwestern University
> www.numis.northwestern.edu
> Corrosion in 4D: www.numis.northwestern.edu/MURI
> Co-Editor, Acta Cryst A
> "Research is to see what everybody else has seen, and to think what nobody
> else has thought"
> Albert Szent-Gyorgi
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20191221/e855f072/attachment.html>
More information about the Wien
mailing list