Forgot to cc the list as well....
You shouldn't have to mess with rsize and wsize if you use TCP, set it to as high as the filer (it should default as high as the filer 32k in nfsv3). The reason you have to mess with it using udp appears to be flow control related and only in mixed speed networks in my experience (gig-e filer to 100mb clients).
What I would do is confirm you are really using TCP, several implementations of the mount command seem to ignore bad options, run tcpdump and copy and file, make sure it's sending tcp packets. On solaris it's proto=tcp, however on linux I think it's just -o tcp. The wierd thing is, if you give it a wrong option under some implementations it returns a prompt, and the mount command even shows that option!
--- Duncan Greenwood duncan.greenwood@btinternet.com wrote:
Try the tech papers at http://www.netapp.com/tech_library/
e.g. http://www.netapp.com/tech_library/3129.html
I played with 9iRAC on RHAS 2.1 against NetApp
last
year and seem to recall we needed to experiment with rsize and wsize to get decent i/o performance, but that was over GbE.
See the tips at:
http://nfs.sourceforge.net/nfs-howto/performance.html
hth
D
----- Original Message ----- From: "Geoff Hardin" geoff.hardin@dalsemi.com To: "Haynes, Tom" thomas@netapp.com Cc: "toasters" toasters@mathworks.com Sent: Friday, March 05, 2004 7:37 PM Subject: Re: Linux / NetApp / Oracle question
I've had several people ask for more details
about
the configuration:
Network: 100 Mb switched network full duplex
Filer: F760C with ONTAP 6.4.1
Database Server: RHEL Kernel 2.4.21-9.EL 1 GB of RAM; (we're trying to get that increased
because we have seen
some swapping going on) 1 Intel 2.4 GHz Xeon processor Mount options:
rw,fg,hard,nointr,retrans=2,tcp,vers=3,timeo=600
We are using NFS over TCP by the recommendation
of
NetApp and because we
saw data corruption when we were using NFS over
UDP. I am looking into
some of the mount options for Linux; one
document
says to turn noac off,
another says to turn it on. I haven't seen any
reference to a
forcedirectio option, and I know that asynch I/O
is not supported in NFS
for Linux (yet?).
If you have any other questions, let me know and
I'll do my best to
answer them. Thanks to everyone who has
responded
so far!
Geoff
Haynes, Tom wrote:
Geoff,
You should have mentioned what type of filer
and
what version of
OnTap you were running.
Thanks, Tom
Hey Toasters, I have a situation that has come up where
we
are testing Oracle 9i
on a 2.4 GHz Intel Xeon system with Red Hat
Enterprise Linux 3.0 as the
operating system. The database itself resides
on a NetApp filer. We
have run through the normal problems,
switching
to NFS over TCP rather
than UDP, no asynch I/O, not as much like
Solaris as we expected, but
now we're still not seeing the level of
performance that we had
expected. Our test database run on a Sun 280R
takes about 2 hours; the
same test run on our Linux system takes about
2:40. Definitely not the
results we had expected. I was just wondering if anyone out there
has
done something similar
and has any tuning recommendations. Oracle
tells us that it should work
fine on Linux, and we know our other databases
work fine running off
filers, so that leaves us with a solution that
should work, but doesn't.
Any recommendations, white papers, or RTFM
(as
long as you indicate
which manual), would be greatly appreciated.
Geoff Hardin UNIX System Administrator Dallas Semiconductor geoff.hardin@dalsemi.com
Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.593 / Virus Database: 376 - Release Date: 20/02/2004
Do you Yahoo!? Yahoo! Finance: Get your refund fast by filing online. http://taxes.yahoo.com/filing.html
__________________________________ Do you Yahoo!? Yahoo! Search - Find what you�re looking for faster http://search.yahoo.com