is TMPFS still highly experimental?

[Olivier Smedts]

Le lundi 3 octobre 2011, Xin LI <delphij at gmail.com> a écrit :
> On Sun, Oct 2, 2011 at 3:33 PM, Ivan Voras <ivoras at freebsd.org> wrote:
>> On 02/10/2011 07:27, Xin LI wrote:
>>
>>> The problem Ivan have asserted was not confirmed by anyone who have
>>> swap configured properly.  Gleb have pointed out that it might be
>>> related to a series of integer overflow by the way (he have also fixed
>>> a lot of tmpfs issues by the way).
>>
>> Well, instead of guessing I can point you to the way I got the original
>> situation - you said you have ZFS configured so it would be easy for you
>> to check.
>>
>> You should to something like this:
>>
>>        - configure your system to the best of your abilities (but post
what
>> you did different from the defaults)
>
> It's mostly "normal" configuration -- 6GB RAM, 14GB swap, ZFS as /usr;
>
> The following sysctl tweaks were done to make postgresql start (these
> are not scientific, just large enough to make it work):
>
> kern.ipc.shmmax=4294967296
> kern.ipc.shmall=1048576
>
>>        - install postgresql (8.4+, but I don't think the version is
>> important), configure it so it gets half the system memory or 2/3 the
>> system memory for its shared_buffers.
>
> Configured to half of system memory (3072MB).
>
> # ipcs -a
> Message Queues:
> T           ID          KEY MODE        OWNER    GROUP    CREATOR
> CGROUP                 CBYTES                 QNUM
> QBYTES        LSPID        LRPID STIME    RTIME    CTIME
>
> Shared Memory:
> T           ID          KEY MODE        OWNER    GROUP    CREATOR
> CGROUP         NATTCH        SEGSZ         CPID         LPID ATIME
> DTIME    CTIME
> m        65536      5432001 --rw------- pgsql    pgsql    pgsql
> pgsql               6   3295936512         2130         2130 10:57:12
> 11:09:17 10:57:12
>
> Semaphores:
> T           ID          KEY MODE        OWNER    GROUP    CREATOR
> CGROUP          NSEMS OTIME    CTIME
> s        65536      5432001 --rw------- pgsql    pgsql    pgsql
> pgsql              17 11:09:25 10:57:12
> s        65537      5432002 --rw------- pgsql    pgsql    pgsql
> pgsql              17 10:57:12 10:57:12
>
>>        - install pgbench
>
> Done.
>
>>        - initialize pgbench so that the database definitely is larger
than the
>> entire memory you got (NOTE: THIS IS NOT A CONTRIVED TEST - lots of
>> databases in practice are larger than the RAM in the server).
>
> Initialized with pgbench -i -s 500 pgbench.
>
>>        - 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?

Try reducing the swap size to less than the RAM size. No "configuration
issue", try with some RAM + swap that should fit all.

And please excuse if we're not speaking of the same tmpfs caveat :-)

>
>> 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!
>>
>>
>> (for more info on pgbench, see this:
>> http://www.westnet.com/~gsmith/content/postgresql/pgbench-scaling.htm).
>>
>> _______________________________________________
>> freebsd-fs at freebsd.org mailing list
>> http://lists.freebsd.org/mailman/listinfo/freebsd-fs
>> To unsubscribe, send any mail to "freebsd-fs-unsubscribe at freebsd.org"
>>
>
>
>
> --
> Xin LI <delphij at delphij.net> https://www.delphij.net/
> FreeBSD - The Power to Serve! Live free or die
> _______________________________________________
> freebsd-fs at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-fs
> To unsubscribe, send any mail to "freebsd-fs-unsubscribe at freebsd.org"
>

-- 
Olivier Smedts                                                 _
                                        ASCII ribbon campaign ( )
e-mail: olivier at gid0.org        - against HTML email & vCards  X
www: http://www.gid0.org    - against proprietary attachments / \

  "Il y a seulement 10 sortes de gens dans le monde :
  ceux qui comprennent le binaire,
  et ceux qui ne le comprennent pas."