Calculating IOPS Used By EBS Volume

Calculating IOPS Used By EBS Volume

How To Calculate IOPS Used By EBS Volumes In AWS?

For GP2 volumes:

Data is collected every 5 minutes, so there is no data available for every minute.
For example, if you see Max of volumeReadOps and volumeWriteOps for a period of 5 mins and 1 min, the value would remain same. If you also see Sum of volumeReadOps and volumeWriteOps for a period of 5 mins and 1 min, the value would again remain same. Below is an observation of this:

statistics - maximum
period - 1 min
value - 285491

statistics - maximum
period - 5 min
value - 285491

statistics - maximum
period - 15 min
value - 285491

statistics - sum
period - 1 min
value - 285491

statistics - sum
period - 5 min
value - 285491

statistics - sum
period - 15 min
value - 830694

So, If you want to calculate the IOPS consumed here, then, divide the IOPS with 300 (that is 5*60) in case you are taking max/sum of 1 or 5 min. The calculation would be 285491/300 = 951.63
If you are considering sum of 15 min, then divide the IOPS with 900 (that is 15*60) which is 830694/900 = 922.99

Above values are being considered when the IOPS were happening on peak. You may consider the data for last 2 week or may be 3 week and choose the window for the peak duration and for that peak calculate the IOPS.

For IO1 Volumes:

The data for IO1 type of volume is sent per minute by default which means you should calculate the IOPS used by your volume by dividing the IOPS by 60 if you choose 1 min duration for sum/max. If you are choosing sum/max for 5 min duration then you should divide the IOPS by 300 (that is 5*60).

Inference:

If calculated IOPS is equal or approximately equal to the IOPS given by you and the volume is constantly running at this IOPS, then, probably your disk may be a bottleneck for the performance of your application. If you see any performance issue and constant utilisation of IOPS then you might need to modify the volume with more IOPS. Here I have used "probably" as this could be one of the reason for low performance but not definitely. There could be some other bottlenecks too like CPU, so do your debugging and research before making a conclusion.