There are multiple ways to measure cifs latency.
Raw data can be measured from the command line using stats, such as:
system> stats show cifs:cifs:cifs_latency
cifs:cifs:cifs_latency:1.92m
Stats can be run in a couple of different ways:
–Once, where current counter values are displayed
stats show
–Repeating, in which counter values are displayed at a fixed interval
stats show –i 1
–Period, in which counters are gathered over a single period of time and then displayed
stats start then stats stop
There are a number of commands in Data ONTAP where statistics can be collected over
a period of time like this. I always recommend that these commands are run from a script
of the general form:
start/zero
sleep 60
stop/output
The reason I give for this is: There's lies, damn lies, and statistics ! When doing performance
analysis I'd rather be looking at real statistics. Using a script you're always looking at the
same time interval.
Counters measured by stats can also be measured by using SNMP. If CIFS is a big part
of your environment you may want to set up an SNMP poll just for CIFS latency.
You can measure latency overall, or by volume for finer grain monitoring.
There's also the advanced priv smb_hist command you can use, which shows CIFS response
delay distributions for all operations.
At your service,
Eugene Kashpureff