Total IOPS

[tyrone_owen_1]

Hi,

I want to get a breakdown of the total read/write/other IOPS over a specific period, say 24 hours or a week for example. OCUM only seems to provide avg/min/max - is there a way to get the total performance data I want a different way?

Thanks

[Mjizzini]

You will find the QOS commands useful to what you are looking for.

QOS commands

[tyrone_owen_1]

Thanks very much for the reply.

Doesn't the qos command provide current statistics as opposed to a total over a 24 hour/week period?

[robverhoeven · Accepted solution]

Hi,

I think NetApp Harvest or NABox can help out here. Check https://nabox.org for more details.

Cheers,

Rob

[tyrone_owen_1]

Thanks Rob, that looks interesting - why would I use it over OCUM?

Looking at the screen shots I see avg/max/min IOPS and not total IOPS broken down into r/w/o for a given period.

[pedro_rocha]

I normally prefer to investigate performance related events through Grafana (NAbox) graphs. AIQUM is very nice though.... I think it's more a personal preference because once you get used to AIQUM, you do almost the same tasks in terms of analyzing graphs and etc.

If you don't ever used NAbox, you should give it a try. It looks faster and more fluid to pin point those IOPS/latency/Throughtput events using graphs and different time frames.

Regards,

Pedro.

[tyrone_owen_1]

Hi,

I'm not investigating a performance issue.

Just to restate the request: is there any way for me to establish the Total read/write/other IOPS for a defined period? Preferably using native tools, e.g. CLI, OCUM....

Thanks

[pedro_rocha]

The below image is from AIQUM 9.7. Is this what you want?

If so, in AIQUM you can do it for the SVM, volume, LUN... not for a cluster or node. I think that you can do it for the node/cluster on NAbox.

Or do you want it in a text format?

[tyrone_owen_1]

Thanks for the reply.

So from the diagram, what is the total IOPS for the 16th November?

[pedro_rocha]

Hello

you'd need to hover the mouse to that date in the histogram, or choose the specific date

[tyrone_owen_1]

But it doesn't give you the total IOPS for a 24 hour or weekly period. It gives you the total IOPS for a specific time  which is not what I am asking for.

Total IOPS broken down into read/write/other for a period of time, e.g. 24 hours, a week, etc. And  not total IOPS for a specific time

[pedro_rocha]

Hello my friend,

Is this what you want?

Regards,

Pedro.

[tyrone_owen_1]

Hi Pedro,

That looks like it could be it - that's not a native NetApp tool though is it?

Thanks

[pedro_rocha]

Hi,

Unfortunately I do not know... maybe someone else here could say.

You could also open a support ticket to check.

As I said before, nabox worth the try...

Good luck!

[ttran]

Hi Tyrone_owen_1,

The screenshot @pedro_rocha provided is captured from a NAbox deployment that @robverhoeven suggested earlier in the thread. NAbox leverages NetApp Harvest, Graphite, and GrafanaLabs to provide a customizable holistic view. Harvest is the data collector, Graphite is the metrics database, and Grafana provides the dashboards.

There is a short intro video, screenshots and everything you need to setup NAbox .

Regards,

Team NetApp

[tyrone_owen_1]

Just to confirm then, as no-one has said specifically, there is no way to do this with native NetApp tooling?

[Alma]

In OCUM there is no presentation of the total IOPS for a period of time - like 24 hours. However in the zoom view shown above you will get the average of the IOPS over the period of time. That can be used to calculate the total IOPS as (average IOPS * number of seconds in the period of time of interest). This number will grow quickly with the length of the period of time. For this reason, i.e., avoid large numbers, OCUM shows only averages of IOPS.

For 72 hour period there is an inventory page that lists all the volumes and displays the average IOPS - again from there one can calculate the total IOPS for 72 hours.

[tyrone_owen_1]

Thanks for the response.

Unless I am missing something, your calculation assumes that the average IOPs for every second during a given period is absolutely consistent, which in many systems I'm sure this is the case.

[Alma]

the average you see in the zoomed graph - together with min max and 95th percentile - is the average of the IOPS value every measurement point in the period of time - so the differences in IOPS in that period of time are already captured by that average number

average = (sum of IOPS every point in time) / (total points in time)

Since you are looking for sum of IOPS across all points in time, average can be used to stretch that number across every second of the period of time of interest.