is TMPFS still highly experimental?

[Ivan Voras]

On 3 October 2011 20:37, Xin LI <delphij at gmail.com> wrote:

> It's mostly "normal" configuration -- 6GB RAM, 14GB swap, ZFS as /usr;

> Initialized with pgbench -i -s 500 pgbench.

Ok, this should be enough ~~ 7 GB database on a 6 GB machine.

>>        - run pgbench & observe the results.
>
> pgbench -t 20000 -c 64 -S pgbench
>
> Can't seem to reproduce:
>
> # df -Httmpfs
> Filesystem    Size    Used   Avail Capacity  Mounted on
> tmpfs          10G     69k     10G     0%    /tmp
>
> Any suggestions?

Can't think of anything to add :) I don't think increasing the
database size would affect it much.

Comparing what you did to my original report at
http://www.mail-archive.com/freebsd-current@freebsd.org/msg126491.html,
it seems like that bug is gone!

But my machine was one of the bigger ones (for that time) - AFAIK 24
GB RAM and many cores (16?). Maybe there is some race condition which
is exposed on large machines.


>> This should create really bad contention problem for memory between
>> postgresql and ZFS, which should manifest itself in tmpfs shrinking to 0
>> bytes free.
>>
>> If you don't get this problem then great, it might be solved!


I don't have the equipment now to test it myself but I agree that
based on what you posted it looks solved.