[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Ordered vs. journal real-worl performance



Hi,

On Mon, Nov 25, 2002 at 09:48:22AM -0500, Rechenberg, Andrew wrote:
 
> Maybe I should've started a new thread with this question (it was in the
> /proc/sys/vm/bdflush thread), so I am now :)
> 
> According to tests performed for this article:
> 
> http://www-106.ibm.com/developerworks/linux/library/l-fs8/
 
> "ext3's data=journal mode is incredibly well-suited to situations where
> data needs to be read from and written to disk at the same time."  This
> is the situation that exists on my box.

It's particularly suited when you have applications performing on-disk
transactions --- ie. when they are using O_SYNC or fsync() to flush
data to disk.  The data-journaling allows the data to be written
sequentially in the log.

However, _eventually_ it will still need to write the data elsewhere
on disk (unless it rapidly gets deleted), so the effect is really to
mitigate the cost of the application synchronisation, rather than
anything else.

So it really depends very much on what the application is doing
whether or not this makes much of a difference.

Cheers,
 Stephen





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]