Some things to *try*

1. Make sure (2.6) patch 105720-08 or later is in place.

2. Make the following changes in /etc/system on solaris 2.6 clients: set ncsize=8000 set nfs:nfs3_max_threads=24 set nfs:nfs3_nra=10 set nfs:nfs_max_threads=24 set nfs:nfs_nra=10

3. Turn large packet support on the filer: oprions nfs.udp.xfersize 32768

If you must use NFSv2, the changes above may also help, but obviously the spec for V2 won't allow greater values than 8K for rsize/wsize.

4. There may be a few other things to try, but for the most part, these are the majors.

--tmac

4. unmount from the filer and remount with default values. You should end up with a vers3 mount with 32K sizes for rsize & wsize.

"Michael S. Keller" wrote:

Lewis,

Necessary to yell?

I use what I received. I am required to stick with it for a while yet.

Solaris 2.6 run against F760s with ONTAP 5.3.4 performs more slowly local storage would. If you can refute this, then tell me how to fix my installation.

I _must_ run NFSv2 to get decent performance, and ONTAP limits NFSv2 packets to 8K. So I must suffer with small transfers. I don't yet have enough storage on these filers to build another test system with Solaris 7 that can handle the load. I'm stuck in the middle without resources to test adequately that Solaris 7 would handle the load better.

You make a blanket statement without supporting facts.

"Kirschner, Lewis" wrote:

...

ABSOLUTELY NOT TRUE DESPITE WHAT THE GENTLEMAN SAYS BELOW!!!!!!!!!!!!!!!!!!!

WE'VE GOT SEVERAL THOUSAND CUSTOMER WHO WILL TELL YOU OTHERWISE!!!

-----Original Message----- From: Yael Hellmann [mailto:yael_hellmann@yahoo.com] Sent: Wednesday, March 22, 2000 2:59 PM To: michael.alvarado@netapp.com; toasters@mathworks.com Subject: RE: nas storage

You are still limitted by the network and netapp will not be able to compete with products which provide direct disk attachment. Products in mind are SANergy and CXFS. Below is a message from a user who supports my assumtion that you cannot get local performance over a network link.

"From: Michael S. Keller [mailto:mkeller@mail.wcg.net]

Sent: Tuesday, March 21, 2000 10:56 AM To: toasters@mathworks.com Subject: Re: nas storage

I do not get performance on par with my perception of local disk storage. Would you care to elaborate on what you mean by "on par with local disk storage"?

When I run Usenet news spools from Solaris 2.6 on a NetApp filer, I must throttle NFS to get it to work reliably. "

--- "Alvarado, Michael" michael.alvarado@netapp.com wrote:

...

Network Appliance's current customer experience is that they do get performance on par with local disk storage already.

Our current plans do not include any interoperation with the SANenergy products.

-----Original Message----- From: Yael Hellmann [mailto:yael_hellmann@yahoo.com] Sent: Monday, March 20, 2000 6:02 PM To: toasters@mathworks.com Subject: nas storage

I am currently investigating the market for nas storage. What is netapp plans in providing a solution which will deliver local disk performance for nas storage? Does netapp have plans for products with SANergy functionality or performance?

-- Michael S. Keller, Technology Solutions Consultant, Sprint Enterprise Network Services On loan to Williams Communications Group Voice 918-574-6094, Amateur Radio N5RDV

"Timothy A. McCarthy" wrote:

Some things to *try*

1. Make sure (2.6) patch 105720-08 or later is in place.

# ls -ld /var/sadm/patch/105720* drwxr-x--- 2 root other 512 Apr 14 1999 /var/sadm/patch/105720-06 drwxr-x--- 2 root other 512 Sep 8 1999 /var/sadm/patch/105720-08 drwxr-x--- 2 root other 512 Feb 9 12:28 /var/sadm/patch/105720-11

Some time has passed since I installed that patch. No help for my situation.

2. Make the following changes in /etc/system on solaris 2.6 clients:

set ncsize=8000 set nfs:nfs3_max_threads=24 set nfs:nfs3_nra=10 set nfs:nfs_max_threads=24 set nfs:nfs_nra=10

I already explored changing these NFS options, to no avail. I have them at defaults again. Another stated that they have no effect on client operations. In my case, I agree.

This host has 1024MB RAM. ncsize would therefore default to 17498. Why would _reducing_ it improve client performance?

3. Turn large packet support on the filer: oprions nfs.udp.xfersize 32768

If you must use NFSv2, the changes above may also help, but obviously the spec for V2 won't allow greater values than 8K for rsize/wsize.

nfs.udp.xfersize 32768 (shared in takeover)

Also set long ago.

4. There may be a few other things to try, but for the most part, these are the majors.

--tmac

4. unmount from the filer and remount with default values. You should end up

with a vers3 mount with 32K sizes for rsize & wsize.

And my news spools will crawl, interactive performance deteriorates and service times in iostat skyrocket. Even with smaller packet sizes and UDP. I _must_ run NFSv2 to get tolerable performance.

Chris Thompson wrote:

It's also true that some performance problems, especially on unlinks, could be dramatically alleviated by decreasing ncsize down to less than 30000 if the default came out larger than that. That's with the software as it was a year ago; I haven't taken off the screwdown to 17500 on our 768 MB machines that I put in at that time, so recent patches may have improved things.

Unlinks should be infrequent, since the news servers run bCandid's software, which uses large database-like files, instead of one file for each article, as some other news server software does.

Reducing it right down to 8000 does sound overenthusiastic to me.

[Meta-note: I find myself talking about Solaris on a NetApp mailing list! I'm going to get into as much trouble as I did for talking about NetApp's on comp.unix.solaris recently...]

Solaris and its interaction with ONTAP. It's on-topic to me.