We currently have a couple of Netapp 760 each with a 7x18gb shelf in a cluster config. The best performance we can get out of this is about 10MB writes.
So we decided to test out what a gig card would do for us and the best we could get was 16MB writes and 30MB reads.
In terms of testing the number of disks in a volume we saw an increase in performance by using a 7disk volume as opposed to a 3 disk volume, but surprisingly we didn't see a substantial increase when we went to 14 disks in a volume.
So have any of you been able to test and justify using 14 disks as opposed to 7 disks in a volume?
Have any of you been able to get better than 16MB writes?
It also doesn't seem to benefit by having one shelf on one FC-AL card and another shelf on another FC-AL card?
It seems the entire performance from what I can tell is limited by the nvram size. (16MB, half of 32MB nvram), thoughts?
We have determined its not the gig cards in our suns that is limiting it as we can get 100MB sun gig card to sun gig card.
Thanks
Art Hebert Email: art@arzoon.com mailto:art@arzoon.com Arzoon IT Dept. Phone: 650 522-6061 Mobile: 650 444-8667
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds. Copies sucked, after 7 minutes I killed it and only 6mb had been copied. Turned on nfs over tcp on the filer, remounted, and viola; reads/copies rule again.
~JK
Art Hebert wrote:
We currently have a couple of Netapp 760 each with a 7x18gb shelf in a cluster config. The best performance we can get out of this is about 10MB writes.
So we decided to test out what a gig card would do for us and the best we could get was 16MB writes and 30MB reads.
In terms of testing the number of disks in a volume we saw an increase in performance by using a 7disk volume as opposed to a 3 disk volume, but surprisingly we didn't see a substantial increase when we went to 14 disks in a volume.
So have any of you been able to test and justify using 14 disks as opposed to 7 disks in a volume?
Have any of you been able to get better than 16MB writes?
It also doesn't seem to benefit by having one shelf on one FC-AL card and another shelf on another FC-AL card?
It seems the entire performance from what I can tell is limited by the nvram size. (16MB, half of 32MB nvram), thoughts?
We have determined its not the gig cards in our suns that is limiting it as we can get 100MB sun gig card to sun gig card.
Thanks
Art Hebert Email: art@arzoon.com mailto:art@arzoon.com Arzoon IT Dept. Phone: 650 522-6061 Mobile: 650 444-8667
On Tue, 30 Apr 2002, Jeff Kennedy wrote:
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds.
200 megabytes in 2 to 4 seconds, meaning 50 to 100MB/sec writes?! What kind of Sun and what kind of Netapp?
420/280/4500 for Sun. 880 filer for NetApp. 17 drive volume with 2 raid groups. Foundry BI8k in the middle.
~JK
Brian Tao wrote:
On Tue, 30 Apr 2002, Jeff Kennedy wrote:
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds.
200 megabytes in 2 to 4 seconds, meaning 50 to 100MB/sec writes?!What kind of Sun and what kind of Netapp?
Brian Tao (BT300, taob@risc.org) "Though this be madness, yet there is method in't"
+-- "Jeff Kennedy" jlkennedy@amcc.com once said: | > On Tue, 30 Apr 2002, Jeff Kennedy wrote: | > > | > > I just had a similar problem that my SE solved for me. For me it was | > > Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m | > > mkfile in 2-4 seconds.
I must have missed the first post on this - this just caught my eye. We have had a "Sun/NetApp nfs over udp problem" that we have yet to resolve (very slow writes). We're currently running over tcp as a stop-gap. We appear to have a very similar setup (Netapp, Sun, Foundry) to your shop.
What was the solution in your case? Can you share a pointer?
Thanks,
Oz
As I think back on it, 4 seconds was probably the starting point for the 200mb file write, not 2. I was switching between 100m and 200m file sizes to see if there was difference (to rule out NVRAM blocks).
It still blazed though.
~JK
Brian Tao wrote:
On Tue, 30 Apr 2002, Jeff Kennedy wrote:
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds.
200 megabytes in 2 to 4 seconds, meaning 50 to 100MB/sec writes?!What kind of Sun and what kind of Netapp?
Brian Tao (BT300, taob@risc.org) "Though this be madness, yet there is method in't"
On Tue, Apr 30, 2002 at 07:32:26PM -0400, Brian Tao wrote:
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds.
200 megabytes in 2 to 4 seconds, meaning 50 to 100MB/sec writes?!What kind of Sun and what kind of Netapp?
I think his mkfile was actually open(); truncate(); close(); which writes close to ligthspeed fast with any file size. ;)
p.
On Tue, Apr 30, 2002 at 07:32:26PM -0400, Brian Tao wrote:
I just had a similar problem that my SE solved for me. For me it was Sun's/NetApps nfs over udp issue. On writes I got nice speed, 200m mkfile in 2-4 seconds.
200 megabytes in 2 to 4 seconds, meaning 50 to 100MB/sec writes?!What kind of Sun and what kind of Netapp?
I think his mkfile was actually open(); truncate(); close(); which writes close to ligthspeed fast with any file size. ;)
Solaris mkfile(1m) will do that (well, roughly) only if -n is specified. By default it writes the whole file with zeros, with the intent to force allocation of all blocks. That's what it was designed for: specifically, initialising swap files.
Chris Thompson Email: cet1@cam.ac.uk
opposed to a 3 disk volume, but surprisingly we didn't see a substantial increase when we went to 14 disks in a volume.
So have any of you been able to test and justify using 14 disks as opposed to 7 disks in a volume?
Art,
Was the 14-disk setup all one RAID group or was it two 7-disk RAID groups? I believe you'd want two 7-disk RAID groups, however 16MB/sec writes on a F760 is probably typical.
You might examine the configuration Netapp uses in their Spec tests (http://www.spec.org).
/Brian/
-
Art Hebert -
Brian Long -
Brian Tao -
Chris Thompson -
Jeff Kennedy -
Ozzie Sabina -
Piotr KUCHARSKI