Adaptive QoS | How the hell is IOPS/TB validated, achieved and managed ?

[ASH2017]

Hi,

Can someone in the community or from NetApp professional services or NetApp AQoS development team provide some insight into the AQoS maths and its working.

For example:

• An application database play area is: 20gb volume with constant 200 IOPS requirement, how does this fit into the SLO Model which defines IOPS per TB?
• If the application read/write ratios change - from write to constant reads - How does it fit in to SLO mdel?
• Per TB is this for thick provisioned or Thin provisioned volumes?
• What if application is assigned 1TB thick but its never going to use that much space?

I was thinking, like a goold old school of thought - It depends how many spindles you have with cache or SSD for that volumes, and then it does not matter how much space you have given to the application, it will always have a contantst IOPS, and you should be able to control it by enforcing the IOPS bandwidth?

Some information with examples would be really appreciated.

Cheers!

Thanks,

-Ashwin

[ASH2017]

Hi,

Anyone from AQoS product team could help with this query ?

Thanks,

-Ashwin

[ASH2017]

Hi,

Still waiting on this this query ? NetApp AQoS Gurus ?

Thanks,

-Ashwin

[ASH2017]

Hello,

I am still awaiting to hear from NetApp ont his.

In the mean time, can any of the customers presently deploying or have deployed SLO IOPS/TB model talk about thier experience ?

Thanks,

-Ashwin

[ASH2017]

Hi,

Can NetApp AQOS product team provide some insight on this topic please ?

Thanks,

-Ashwin

[AlexDawson]

Hi there,

Sorry you haven't got an answer to this yet - I would suggest it would be appropriate to raise a ticket with our support organisation to ensure a response from appropriate resources, or talk to your contacts from our Service Design Workshop.

As a reminder from our documentation, we do not suggest or support using the SLO WFA plugin without having first having had a Service Design Workshop consultation with our architecture team. 

Hope this helps.

[shenoy]

Hello Ashwin,

Here below are answers to your questions. Let us know if you need further help or clarification.

An application database play area is: 20gb volume with constant 200 IOPS requirement, how does this fit into the SLO Model which defines IOPS per TB?

Each “Storage Service Level” has a minimum number of IOPS allocated.

For example, with a 20gb volume using the “Value Service Level”, 75 IOPS are allocated.

A-QoS policies accommodate small volumes (GB size) by means of “Absolute Minimum Value” field. This is a fixed IOPS number. This is meant to provide some performance level and not stifle performance completely. One could use custom A-QoS policies to set different values from default policy, if required.

A-QoS is supported natively in ONTAP 9.3 which supports creating custom A-QoS policies using CLI.

If the application read/write ratios change - from write to constant reads - How does it fit in to SLO mdel?

Using the example above with a constant reads workload, up to 75 reads can be processed by ONTAP per second.  

Per TB is this for thick provisioned or Thin provisioned volumes?

The IOPS allocation is the same for thin and thick provisioned volumes.

What if application is assigned 1TB thick but its never going to use that much space?

IOPS are allocated for the empty volume and IOPS allocation increases based on the amount of data stored in the volume.

Thanks for your interest in ONTAP QoS and Adaptive QoS. Much appreciated.

Best,

--Pradeep

[ASH2017]

Hi Pradeep,

Thanks a lot for sharing the information. I really appreciate it. To be honest with you, it will take some time before we fully understand how it works. Not sure, if this is our lack of understanding or the product's lack of explanation.

Sorry to bother you with my follow up question.

If suppose I am attaching SLA 'Performance' to the volume [for sql app], it says - 8 miliseconds, 1024 IOPS/TB. 

Does it mean - I don't have to have a footprint of 1TB to get 1024 IOPS if need be ? Can I know whats the IO size for 1024 IOPS ? If its SQL then are we not doing IOPS  64 KB  to 1MB ?

In other words - 1024 maximum IOPS that can be provided on that tier up to 1 TB, that means even with 2 GB foot print of database volume , I am able to draw 1000 IOPS ?

Thanks,

-Ashwin

[shenoy · Accepted solution]

Hello Ashwin,

Thanks for asking Adaptive-QoS related questions. This concept of IOPS/TB is fairly new and may take a while to understand. Adaptive QoS works well when used in conjunction with NetApp Service Level Manager (NSLM). I'll have a link for details regarding NSLM for you to get familiar with.

https://mysupport.netapp.com/documentation/productlibrary/index.html?productID=62384

NetApp also offers Service Design Workshop (SDW) for customers to help with storage QoS and Adaptive QoS in particular in their environments.

For your below example there needs to be enough performance headroom to handle IOPS, throughput, latency requirements and storage needs to be placed accordingly. SLA of 8 milliseconds may require AFF platform. For 2GB volume it’s a fixed # of IOPS allocated which is called "Absolute Minimum IOPS". The IO size is application dependent, although, for calculation purposes we use 32K average size.

Hope this helps.

Cheers.

--Pradeep

[ASH2017]

Hi Pradeep,

Thanks a lot for your answeres. I really appreciate it. Getting timely responses on the forum thread is always very encouraging.

I guess it willl take some time, but will get there.

Our Company has already engaged NetApp professionals for helping us understand this feature, but until we fully ingest it we will continue to ask questions.

Thanks,

-Ashwin