My answer today is sponsored by the NTAP Advanced
ONTAP class. This class covers how to measure items
like the ones suggested below. Check out the NOW site
for times, places, and costs..it's worth it!
Now to the answer:
To view NVRAM stats, you use the wafl_susp command. To
find this command, you'll need to be in rc_toggle_basic
or priv set advanced mode.
To zero out the stats, use:
filer*> wafl_susp -z
Then run your test.
Then to view the stats, run:
filer*> wafl_susp -w
You'll get screens full of info but the really important
ones are the following 3:
cp_from_timer
cp_from_log_full
cp_from_cp
The first 2 are good, the 3rd one is bad. cp_from_timer is
the number of times your NVRAM was flushed because half of the
NVRAM was not filled in 10 seconds. cp_from_log_full is when
you filled up half of your NVRAM and flushed it before the 2nd
half of your NVRAM filled. Again, not a big deal. The 3rd one,
cp_from_cp means that you filled the 2nd side of your NVRAM BEFORE
you finished flushing the first half. This is bad. It's okay to
have a few of them, but if they get to be a signifcant percentage
of your cps, you're either over-running your NVRAM, or there's a
problem with your disk subsystem and it's not writing out the data
very quickly (like maybe your volume is 99% full).
Now individual disk utilizations are found with statit. This is similar
to wafl_susp in that you start it, then you stop it and print out the
results. The 2 command for statit are
filer*> statit -b
to start collecting
filer*> statit -e
to end it and dump the data.
You'll find how busy your individual disks are that way, plus a BUNCH of
other things, but that's probably the most important.
But before I'd even go down these paths, I'd be watching my sysstat's very
carefully. sysstat 1 is your friend in looking at performance. It can give
you some really good clues.
Also check netstat, ifstat, and the other networking commands for errors.
Check your switch ports for dropped packets. Things like flow control settings
(a good thing, I suggest full flowcontrol on send and receive) can also mess
up Gbit interfaces.
Anyway, this is probably all I can practically jam into a single message and
have it make sense. Of course, the Global Support Centers at NTAP are always
here and can help your diagnose these types of problems as well.
Hope this helps.
-- Adam Fox
NetApp Professional Services, NC
adamfox(a)netapp.com
> -----Original Message-----
> From: Flavio E. Cardoso [mailto:flavio.cardoso@interchange.com.br]
> Sent: Thursday, July 12, 2001 11:46 AM
> To: toasters(a)mathworks.com
> Subject: Benchmarks
>
>
> Hi Srs.
>
> I´m looking for benchmark infomation about FILER 760, mainly about
> NVRAM and disk´s operations (reading and writing).
> We´ve got the Filer and Servers on a Gigaethernet network but when
> we see on switch statistics the thoughtput is slowly than a
> fastethernet.
>
> Flavio Cardoso
>