In regard to: [Networker] NSR76 SP4 & Optimized disk reading ..., FAIDHERBE...:
While previous releases used fixed 64 KB blocks while reading files from
a client, NetWorker 7.6 SP4 introduces an intelligent algorithm which
chooses an optimal block size value in the range of 64 KB and 8 MB based
on the current read performance of the client system. This discovery
occurs during the actual data transfer and does not add any overhead to
the backup process, while potentially significantly increasing disk read
performance.
I mentioned this in response to a post from Eddie Albert regarding
performance improvements in NW 8.0. This feature is a backport from 8.x.
Our windows guys have done significant testing of the 7.6.4 client and
its block size tuning, and found that the algorithm doesn't auto-tune as
much as we would like. On some of our SAN storage, it chooses a higher
block size (1 MB is the largest we've seen it auto-tune), but on most
of the SAN storage it just stays with 64K.
We have, however, seen backup improvements by manually setting it to a
larger value. What's weird is that when we force it to be e. g. 1 MB,
tools like Windows Xperf report that it's actually using an 8 MB block
size. It's like there's a factor of 8 when setting it manually.
Did someone notice improvements in backup speed thanks to new read algorithm ?
Manually setting it to larger values has cut as much as 40% off the backup
time for some of our SAN volumes. We haven't deployed it to all our hosts
yet. We started with some Windows 2008 R2 clients with SAN storage that
were backing up slowly. Increasing the cluster size on the filesystem
and using a larger block size with NetWorker has definitely helped.
We'll probably be rolling it out to other clients in the next month or
two.
Tim
--
Tim Mooney Tim.Mooney AT ndsu DOT
edu
Enterprise Computing & Infrastructure 701-231-1076 (Voice)
Room 242-J6, IACC Building 701-231-8541 (Fax)
North Dakota State University, Fargo, ND 58105-5164
| 