Henrik Nordstrom wrote:
> performance due to even less I/O overhead. A fastfs:ed filesystem will
> have close to the same performance carateristics as a tmpfs filesystem,
> but with slightly higher persistense (surives clean shutdown or soft
> crashes which syncs disks, likely to fail on powerloss).
More on the issue of asyncronous filesystem updates:
Using fastfs on solaris may use somewhat more memory for filesystem
buffers than a jornaled filesystem. This should however be cured by
enabling priority paging.
Linux is not by far as sensitive to poverlosses as a fastfs:ed Solaris
filesystem even thought the filesystem is asyncronous. The reasons to
this are a more advanced swapping/paging system used by Linux which
updates filesystem metadata much more frequently than plain data blocks.
However, using a jornaled filesystem has the strong advantage that there
is no extensive filesystem check after a power failure. Also I beleive
there are fewer kernel memory lock situation which could cause degraded
preformance. Linux seems to degrade somewhat while the updates take
place.
--
Henrik Nordstrom
Spare time Squid hacker
