dan...@zoltak.com:
>   Is there anyway to validate this?

Do you mean that you want to make sure that aufs code is unrelated to
the activity out of aufs? How about adding printk in d_revalidate or
code review?


> > How much RAM does your system have?
>
>   16GB and 8 cores.

Memory large enough.
CPU many enough.
Let's keep in mind about lock contension.

Just to make sure, you can see them all via /proc. Right?



>   Something that I have noticed.
>
>   On the AUFS system if I do:
>   # ps -ef |grep flush
>   root      2893     2  0 Jan24 ?        00:00:00 [kdmflush]
>   root      2902     2  0 Jan24 ?        00:00:00 [kdmflush]
>   root      4639     2  0 Jan24 ?        00:00:04 [flush-253:0]
>
>   lsof -p 4639
>   COMMAND    PID USER   FD      TYPE DEVICE SIZE/OFF NODE NAME
>   flush-253 4639 root  cwd       DIR   0,17      260    2 /
>   flush-253 4639 root  rtd       DIR   0,17      260    2 /
>   flush-253 4639 root  txt   unknown                      /proc/4639/exe
>
>
>   On the non-AUFS system if I do the same:
>   # ps -ef |grep [f]lush
>   root       319     2  0 Jan25 ?        00:00:00 [pdflush]
>   root       320     2  0 Jan25 ?        00:00:00 [pdflush]
>   root      2042     2  0 Jan25 ?        00:00:00 [kdmflush]
>   root      2052     2  0 Jan25 ?        00:00:00 [kdmflush]
>
>   I'm not sure if this has any relevance?

- there is no pdflush on aufs system.
  generally it should exist on aufs system too. please check your
  system again.

- flush-xx:xx exists on aufs system only.
  it is a kernel thread for a block device.
  as long as your kernel is unchanged, it should exist on non-aufs
  system too.
  as you might know, aufs doesn't have its backend block device. so
  generally speaking, such kthread is unrelated to aufs.

If local block device is related to your load average, you may find
something by iostat(8) or something.


>   There is no doubt something odd is happening. The performance
>   difference has to lie with the AUFS kernel module somewhere?

I don't know why you call your load average as "performance".
If you want to know how the performance of your apache server is
changed, I'd suggest you to try benchmarking from http client.

If you want to know about the caching in NFS client, I'd suggest you to
try "time find /home -xdev -ls > /dev/null" or something.


J. R. Okajima

------------------------------------------------------------------------------
Special Offer-- Download ArcSight Logger for FREE (a $49 USD value)!
Finally, a world-class log management solution at an even better price-free!
Download using promo code Free_Logger_4_Dev2Dev. Offer expires 
February 28th, so secure your free ArcSight Logger TODAY! 
http://p.sf.net/sfu/arcsight-sfd2d

Reply via email to