IOPS measurement at process level

[edmund_kueh]

Recently, we were running a Dynatrace Managed POC for a customer in Malaysia. We installed the OneAgent on 4 Windows based hosts in the customer production environment.

On the very last day of the POC, the IOPS demand spiked on one of the hosts installed with OneAgent. The customer were able to trace this problem to OneAgent by using their own measurement tool.

Currently in Dynatrace, from the host level, when we zoomed into the contributing process, we can only view the CPU utilization, memory and retransmission for each of the individual processes.

Hence, it will be great if we can display the IOPS utilization for each contributing processes in a host.

[Karolina_Ruszko]

Hi Edmund,

Thank you for reporting this request for enhancement. We should investigate reasons for the IO spike. One of the reasons could have been extensive logs analysis.

As far as the RFE ins concerned, reporting disk IO per process has been on our radar and it's an enhancement we want to pursue in the near future.

Best,

Karolina

[naveen_morisett]

Is there any update?

[edmund_kueh]

Hi Karolina,

Several suggestions from my side:

When reporting on the IOPS per process, can you look into the following:

1. Provide a breakdown of the read and write IOPS. Some process are very read intensive with very little write and vice versa. So by having the breakdown, we can identify the nature of the process, whether it is read or write intensive.

2. Logical vs Physical IOPS. Some IOPS measurement are very misleading. Instead of reporting the exact IOPS that are hitting the disks (physical), they are reporting on the IOPS that are hitting the OS buffer. (logical) Sometimes the logical IOPS can reach a few thousand IOPS. So we need to be careful with this.

[Radoslaw_Szulgo · Accepted solution]

It's already been there for years.