> From: [email protected] [mailto:[email protected]] On
> Behalf Of Ian Stokes-Rees
> 
> The destination is an Apple X-RAID array (4TB) connected to an Apple
> XServe.  The corresponding ganglia snapshot is here:

There is one thing you can do, which will massively improve performance.
Make sure you have WriteBack enabled on your Xraid controllers.

If you don't have a BBU, you don't want to leave the WriteBack enabled for
long.  But if you have the BBU, it's relatively safe to keep the WriteBack
enabled permanently.

I only know of one truly *good* solution to this problem, which is to use
ZFS.  In ZFS, you have the option to make the filesystem treat sync writes
as async writes, so everything gets buffered in RAM and aggregated into
larger sequential blocks before writing to disk.  This yields a 1-2 order of
magnitude performance boost.  Obviously you wouldn't want to do that in the
long run (like running WriteBack without BBU).  But it's very useful in
situations like this.  Also, "zfs send" directly streams the blocks of the
filesystem, instead of performing filesystem level operations.  So again,
1-2 orders of magnitude faster, and the only catch is, you either receive
the entire filesystem, or none of it.

Unfortunately, ZFS isn't an option on the Xserve running OSX.  Think about
it for next time.

_______________________________________________
bblisa mailing list
[email protected]
http://www.bblisa.org/mailman/listinfo/bblisa

Reply via email to