I saw some anecdotal reference on some web page for LTO-2 block sizes,
so I've been using 256k buffers, and have had no problems backing up or
restoring test data (our new cluster, and the systems it backs up, have
not yet gone live). However...

When we got our new hardware (Sun V480s and V1280s running Solaris 9,
SAN-connected to IBM LTO-2 drives in an ADIC i2000 robot) I did
extensive low-level tests and found that an optimal buffer size was
2048k. My tests were done just using 'dd', writing an 8 GB file of
highly-compressable (at least 4:1) data. Using numbers from iostat,
averaged over 1 minute, I was able to write 95 MB/s to tape, with peaks
up to 100 MB/s. Using smaller block sizes, the results were about 25%
slower.

Since the perf tuning guide says not to use any block size that exceeds
the drive or OS's capabilities, presumably you can use any block size
that does work, so 2048k may be OK on systems similar to ours.
Unfortunately, I don't have much more time to test if NetBackup runs
significantly faster with larger blocks than we're already using.

matt

On Wed, 2003-11-05 at 11:30, Shuster, James wrote:
> Does anybody know the "best practices" buffer size setting for LTO II?
> _______________________________________________
> Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu