RE: netapps experience and performance tool

  • From: "McCartney, Bruce" <BMcCartney@xxxxxxxxxxxxxxxxxxx>
  • To: <Ethan.Post@xxxxxx>, <ntilbury@xxxxxxxxxxxx>, <stevenoyle1@xxxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Fri, 19 Nov 2004 12:00:59 -0700

Netapp has some really nice features, such as snapshots that allow for =
very fast point-in-time recovery in very little time - without a mirror =
of all the data.  It does this by maintaining a list of block changes in =
the netapp by time;  allowing you to take snaps at several points in =
time during a process and quickly return to those point in time.  This =
is really nice in a testing environment, but we had some issues with the =
feature in production that you need to be aware of.  If you create a =
snap and just leave it, you eventually run out of space on the device =
and get oracle errors are somewhat difficult to believe because,  for =
example, you get an OS "i'm out of space, so i can't grow" error on a =
device that has nothing but data files are not autoextend enabled...

We also ended up migrating off netapp for performance reasons on a very, =
very redo intensive OLTP application.  It turned out we reached a max =
write throughput for the filer (everything was fine below this =
threshold) and things degraded quite dramatically.  This write rate was =
primarily due to poor application design (update millions of rows;commit =
every row) but was a real issue. With the filer, OS monitoring is very =
different as the OS sees network traffic; not disk access.  We had: =
netapp come in and look at the filer and it "wasn't busy" - based on =
summary statistics that are fundamentally flawed'; OS gurus look at aix =
and solaris and say it is not being taxed and the network is "not busy" =
- based on summary statistics relative to the 2 Gigabit pipes that are =
fundamentally flawed; and the poor dba sees oracle report timed events =
for log file sync and db file % read approaching floppy disk speeds and =
users whose jobs are not finishing when the need them.

Also, with our version of the filer, i recall that aync I/O was not =

Overall, the netapp was deployed happily in less intense database...


-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx]On Behalf Of Post, Ethan
Sent: Friday, November 19, 2004 11:07 AM
To: ntilbury@xxxxxxxxxxxx; stevenoyle1@xxxxxxxxx; oracle-l@xxxxxxxxxxxxx
Subject: RE: netapps experience and performance tool

If you end up NFS mounting netapp I don't think sar -d works.  I just
recently saw an issue with DBWR bottleneck during checkpoints, appears
to be some sort of IO configuration issue with Oracle/HPUX and NetApp,
this article I think may have provided the solution

I am just a point of contact on the issue so I am not sure.  Anyway, I
think in general provided you get a filer with plenty of disks you can
expect pretty good performance, there are other listers here I know are
running NetApp very successfully with heavy IO demands.  I think the
lesson to be learned is make sure you get spread the IO out across a lot
of disks, requires a large filer and don't skimp on configuration, you
don't want the network to be the bottleneck.

-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Nick Tilbury @
Sent: Friday, November 19, 2004 10:09 AM
To: 'stevenoyle1@xxxxxxxxx'; oracle-l@xxxxxxxxxxxxx
Subject: RE: netapps experience and performance tool

I believe Oracle and Netapps are very much 'in-bed-together' now days
all Oracle in-house platforms
are now hosted by Netapps kit (

I have had some seriously bad experiences with write performance on NAS
units but that was a few years ago
and it was on an extremely cheap bit of kit.

My current thinking is there is a place for NAS but unless it's top of
line kit it's place is not=3D20
hosting an OLTP database.

We currently use a NAS unit for DEV/TEST databases and are in the
process of
implementing a disk-staging
procedure for backups using an ATA Beast NAS.



Other related posts: