>
>
> On Wednesday, July 18, 2001 07:51:22 AM -0400 Danny Cox
> <danscox@xxxxxxxxxxxxxx> wrote:
>
> > All,
> >
> > One other point regarding benchmarks, it's relevancy depends upon the
> > environment: when recovering from a "temporal power anomaly" [;-)] an
> > XFS mount will usually take less than a second. Reiser can take
> > several.
> >
> > Admittedly, I'm using Reiser as it exists on SuSE 7.1, and XFS as of
> > this week's CVS, but an XFS mount was always quick (witnessed personally
> > when I was helping to track down XFS/RAID5 incompatibilities...several
> > times...every hour.... ;-).
> >
> > Yes, it's not the "normal mode", but when the system's down, and the
> > users have their noses plastered on the glass staring at your back, that
> > extra 4 seconds per filesystem can seem an eternity!
> >
> ;-)
>
> XFS does logical logging, which allow them to use a very small log, and
> smaller logs take less time to process.
>
> reiserfs logs full blocks, and the journal code tries really hard to use
> the entire 32MB log before it starts flushing blocks to their real
> locations on disk. So, the reiserfs log replay time should be slightly
> longer than it takes your disk to read and write 32MB (in synchronous
> chunks from 12k to 4MB in size).
And you can make xfs recovery slower by making a bigger log. The larger
log means when you are running the filesystem there are less occasions
when you run into the case where you need to flush metadata to get log
space. The larger log though takes longer to process at recovery time.
Steve
|