Retrieve function performance advisor

[MAXWELL_CROOK]

In the performance advisor in the management console 3.3 I can set a threshold on the aggregate throughput. I right click in the graph, select add threshold, select suggest value, set the duration to be a week then hit the retrieve button. This gives me the min,mean, max and 95% percentile value. Does anyone know where these figures are generated from and can they be directly interrogated from DFM ?

Seems a very long winded way for me to find the IOPS per aggregate....

[rmatsumoto]

I'm not sure how you'd do it with NMSDK, but if you were using CLI it'd be:

95th percentile from the last 600 seconds

dfm perf data retrieve -o filer1:aggr1 -C aggregate:total_transfers -d 600 -m value_at_percentile -P 95

mean value between 1pm-2pm today

dfm perf data retrieve -o filer1:aggr1 -C aggregate:total_transfers -b "2014-11-21 13:00:00" -e "2014-11-21 14:00:00" -m mean 

At the bottom of each output you'll see the output for -m.  

if you want to see the value for all the aggr on a controller, just remove the ":aggr1" part and it'll get the numbers for all the aggrs for that controller.  This works the same way for groups, so if you group filers(or aggrs), then you can plug in the group's object ID after -o and get the values that way.  I do that sometimes when I need to pull numbers for an entire DC.

As far as I know, each counter group for each filer gets its own file in the perfdata directory.  You can verify this by typing dfm perf data list -v.

[MAXWELL_CROOK]

Wow! So these commands are run from the dfm cli on the host where the data is collected that the Netapp Management console reports?

I cant believe someone actually knows how to do this! Thank you so so much. I will compare what I get from DFM with the GUI way and that way cross reference to ensure I'm getting the correct figures.

Fantastic.

[rmatsumoto]

That number after -P does not have to be an integer.  It can get you something like 99.99th percentile if you wanted. The use case for that was when I was asked to quantify how often the latency on a certain application got high, I was able to observe a pretty significant difference between the 99th percentile value versus 99.9th percentile.  It may seem like a small difference, but that's a difference between ~10 minutes and ~100 minutes in a week's time at 1 minute sampling intervals, and people certainly care if your storage performs poorly for 10 minutes vs 100 minutes in any given week.   

[rmatsumoto]

"Wow! So these commands are run from the dfm cli on the host where the data is collected that the Netapp Management console reports?"

I see I never answered you directly on this.  Yes, you run this on the DFM host.  Our DFM is running on a win2k8 box and we typically connect to it over a PS-remote session and the commands work fine.

[MAXWELL_CROOK]

It wasn't an issue, you've given me far far more than I even knew was possible. Its opened another window of knowledge for me, I wish these things where communicated in some of the training i've had. I'm truly thankful for this.