I'm evaluating a NetApp F840C and an EMC Celerra. I'm using a benchmark program called Iozone (http://www.iozone.org/) to test NFS throughput. Are there any other benchmark programs out there anyone would suggest I try?
I'm using iozone like this:
iozone -a -e -f /emc/foo -U /emc
or
iozone -a -e -f /netapp/foo -U /netapp
on a Sun E4500, and the numbers I'm getting are about 30 MB/sec read/write for the F840 and 20 MB/sec write and 25 MB/sec read on the Celerra. Both are connected with Gigabit Ethernet.
The -e option does an fsync/fsflush after every write, and the -U does an unmount/mount of the filesystem between every test. This seems to eliminate the huge throughput numbers I was seeing before, due to filesystem cache. (The E4500 has 4 GB of memory.)
Any suggestions would be appreciated!
Mark
At 05:55 PM 9/25/2001 -0700, Mark Bentley wrote:
I'm evaluating a NetApp F840C and an EMC Celerra. I'm using a benchmark program called Iozone (http://www.iozone.org/) to test NFS throughput. Are there any other benchmark programs out there anyone would suggest I try?
Depends on the type of load you're trying to simulate. There are a number of benchmarks out there that try to simulate different things. For example, postmark tries to simulate creating lots of small, short-lived files like you'd see on an e-mail server. Plain ole 'dd' is pretty good at simulating sequential I/O. You'll find a bunch of stuff at:
http://www.acnc.com/benchmarks.html
In general, however, it's best if you can benchmark the actual application you'd be using the filer with. It's quite difficult to create a simulation that accurately predicts real world performance.
On Tue, Sep 25, 2001 at 05:55:43PM -0700, Mark Bentley wrote:
I'm evaluating a NetApp F840C and an EMC Celerra. I'm using a benchmark program called Iozone (http://www.iozone.org/) to test NFS throughput. Are there any other benchmark programs out there anyone would suggest I try?
You might look at:
http://www.coker.com.au/bonnie++/ http://www.netapp.com/tech_library/3022.html (Postmark from NetApp)
Also, depending on your ultimate application, you might try some of the other programs that simulate specific ones -- those are listed at the bottom of the bonnie page.
A bit more detail on your setup might also yield tuning suggestions at least on NetApps from this list...
- NFSv3? - TCP or UDP mounts? - through a switch? what type? - jumbo frames?
on a Sun E4500, and the numbers I'm getting are about 30 MB/sec read/write for the F840 and 20 MB/sec write and 25 MB/sec read on the Celerra. Both are connected with Gigabit Ethernet.
does the Celerra have NVRAM write-cache? that might be the cause of the NetApp showing better write performance...
I'm interested in a summary of the comparison when you are done...
Thanks, Adi
On Tue, 25 Sep 2001, R.P. Aditya wrote:
A bit more detail on your setup might also yield tuning suggestions at least on NetApps from this list...
- NFSv3?
Yes.
- TCP or UDP mounts?
TCP.
- through a switch? what type?
Yes, a Cisco Catalyst 6509.
- jumbo frames?
No.
Any tuning suggestions?
on a Sun E4500, and the numbers I'm getting are about 30 MB/sec read/write for the F840 and 20 MB/sec write and 25 MB/sec read on the Celerra. Both are connected with Gigabit Ethernet.
does the Celerra have NVRAM write-cache? that might be the cause of the NetApp showing better write performance...
I'm not sure about the Celerra config. EMC has been out here several times checking and double checking the config, so I'd think they have it setup optimally. I haven't actually touched the config personally.
Should the Celerra get better write performance?
Mark
On Tue, Sep 25, 2001 at 07:14:16PM -0700, Mark Bentley wrote:
- TCP or UDP mounts?
TCP.
you might try with UDP; depending on your application that _might_ change things. Do you plan to have several clients hanging off the filer?
Should the Celerra get better write performance?
no idea -- I only have experience with NetApps, however we're doing due-dilligence before buying any new NASes...
Adi
-
Jamey Maze -
Mark Bentley -
R.P. Aditya