Thanks for the clarification.
Anyway, to Dearl Neal's point, you won't see much (if any) difference
between cached and not cached in zfs, until the next release, because
it is fs_flush that does the deed, and that is always called at present.
Incidentally, I have an action item to come up with a better name for
that attribute, as "cached" actually means "don't flush the memory
mapped cache" and "not cached" actually means "DO flush the memory
mapped cache". in both cases, file caching happens, it is just a
question of whether the file cache starts out "hot", with portions of
the files that were just created, or "cold".
Drew
On Apr 28, 2009, at 2:22 PM, johan...@sun.com wrote:
On Tue, Apr 28, 2009 at 12:58:25PM -0700, Andrew Wilson wrote:
I believe that Solaris always converts reads and writes into file
mapped
I/O, so if the cached attribute is missing or set to "false" the VM
pages
involved in that will be flushed, so you will see a small
difference in
performance, but because the files still live in the ARC, it is much
smaller than with UFS, where the files have been completely flushed
from
memory.
UFS reads and writes are converted to file mapped I/O because the VM
is
used to back the file cache. In ZFS, the read and write operations
only
perform file mapped I/O when the vnode has an existing file mapping in
place. If the vnode doesn't, then the only place the data is cached
is
in the ARC.
-j
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
