ONTAP Recipes

ONTAP Recipes: Easily use QoS Max to prevent a FlexGroup from becoming a bully object

ONTAP Recipes: Did you know you can…? 

 

Easily use QoS Max to prevent a FlexGroup from becoming a bully object 

 

Problem:  There is no throttle/mechanism to prevent a FlexGroup from consuming more resource than needed and eventually become a bully.

 

Let’s consider the scenario of a Production and Test workloads for this recipe and see how we are trying to avoid the Test workload from becoming a bully.

 

Solution:

 

Use ONTAP QoS Max/Ceilings to limit the test workloads in both IOPS and throughput. By limiting the test workloads, the production workloads will get their required performance met. QoS Max/ceilings for FlexGroup is a new feature in ONTAP 9.3.

 

It is available using the command line interface (CLI) and application programming interface (API). QoS Max/Ceilings for FlexGroups supports both NFS and CIFS protocols.

 

Problem Example:

Figure 1 shows an ONTAP cluster with a FlexGroup provisioned for production applications and a second FlexGroup provisioned for test applications. Without QoS, the test workloads are interfering with production workloads resulting in slow application response time.

 

Picture1.png

 

Figure 1: FlexGroup used for Production Applications is not meeting its required service level due to interference from Test workloads

 

Solution Example:

Use ONTAP QoS ceilings to limit the test workload both in IOPS and throughput.

 

Using ONTAP command line interface (CLI):

 

1. Create a QoS policy with a ceiling of 500IOPS and 2 megabytes/sec (4K block size).

 

::> qos policy-group create -policy-group test_policy -max-throughput 500IOPS,2MB

 

2.Assign the test FlexGroup to the QoS policy

 

::> volume modify -volume test_flexgroup -qos-policy-group test_policy

 

[Job 40] Job succeeded: volume modify succeeded

 

 

Using OnCommand System Manager:

 

1. Verify the test FlexGroup is limited by QoS and is not impacting the production workload.

 

2. Both the test and production FlexGroup have a load of 2000IOPS with a block size of 4K

 

3. The test workload is QoS limited to 500IOPS and 2MB throughput. Since the offered load to the test FlexGroup is 2000IOPS there is QoS induced latency of ~30 milliseconds.

 

The production workload is not QoS limited and is able to serve 2000IOPS with no QoS induced latency.

 

Picture2.png

 

Figure 2: Use System Manager to verify the production workload is no longer impacted

 

Picture3.png

 

Figure 3: Use System Manager to verify the test workload is QoS limited

 

 

For more information, see the ONTAP 9 documentation center

 

 

 

Forums