namei.unix at gmail.com
Thu Dec 20 12:38:43 CET 2012
On 12/20/2012 07:29 PM, Valerio Pachera wrote:
> We made sure that moving journal to a dedicated device increase
> cluster performance.
> What I didn't get yes is: *what size give to journal?*
> The bigger the better? Or does it need to be a precise size?
> PS: I noticed the the "journal device" make more i/o than the "sheep
> dog" device, something like 20%sheepdog device 40%journal device).
> Thank you
The size of journal file will affect node when to commit the data to
disk in the journal file.
The bigger journal file, you'll have
pros: less rounds of committing
cons: longer time for one committing process (VM will be slowed down in
the process of committing)
To get the balance, I think 256M~512M is good enough.
More information about the sheepdog-users