Guest I/O Profiling – Survey Results

A month ago I posted a survey revolving around guest I/O workload characterization. The survey consisted of five questions (I wish I would have included a sixth) that asked respondents about their experience with profiling the I/O that goes to and from the guest VM to the storage subsystem. Let me tell you why this data is important: simply, knowing the I/O workload helps to architect and maintain virtualized environments, especially with regards to the storage subsystem. That means that knowing the I/O size, what percentage is random vs sequential and response times are extremely important to the performance of a virtual machine and its application.  The idea behind the survey is trying to determine:

  1. Are people generating I/O profiles for their virtual workloads
  2. If they aren’t compiling I/O profiles; why not
  3. What tools are being used by the people that are creating I/O profiles
  4. If their primary tool is vscsiStats, how can it be better

As far as tools are concerned, my primary focus was vscsiStats. If you haven’t used vscsiStats before, you really should; immediately. Cormac Hogan posted a nice ‘getting started’ article for how to get started with vscsiStats. I’ve read through it and I’m able to easily follow and understand the article. However, I’ve used vscsiStats before. When I first started using vscsiStats it wasn’t the easiest tool to figure out (although it’s VERY powerful). I often wondered how many people were missing out on this great tool that provides important data in the planning and O&M of any virtualized environment.

Here are the results from a few questions I asked in the survey that I feel answers some of the aforementioned questions:

  • Do you have a need to profile I/O of guest virtual machines? Such as determining the I/O size or latency that the guest OS sees?

image

88% of people said they have a need to profile guest I/O. I had another question that asked ‘Have you heard of vscsiStats for ESXi?’ Based on that question, 31% stated ‘they’ve heard of it but never used it’ and 62% stated ‘they’d heard of it AND used it’

This led me to another question that focused on just the people that answered ‘they’ve heard of it but never used it’:

  • Why haven’t you used it (vscsiStats)?

image

Here’s the good stuff. 46% said they ‘didn’t know how to use it‘’, 46% said ‘It’s too complicated to use’, and finally, 27% said ‘It’s too hard to interpret the results’.

Ok, so I know what you’re thinking; how did I end up with 119% out of 100%? Easy, the question above allows for more than one answer.

Here’s the last question I asked:

  • Would you use a tool like vscsiStats, or some other tool, if it were easy to understand, easy to use, easy to interpret the results and provided the data you needed?

image

As I’m sure all of you already guessed, 92% of respondents said Yes.

If you talk about any tool that is easy to use and you get what you’re looking for, of COURSE people will say yes they want it and would use it. One goal of the survey was to extrapolate whether there are people who have a need for this type of data, and that being able to easily gather and consume that data in an easy manner is important. I also hope that these results can foster support for taking the awesomeness of vscsiStats and turning it to something that is easier to use and easier to consume.

If you have any experiences (good or bad) with vscsiStats or similar tools please leave a comment below. If there are any other barriers to you using these types of tools, please also put that in the comments.

I’d like to thank each and every person that took the time to fill out the survey. I know how the value people place on their time, and rightfully so. Your time is very much appreciated. Finally, what all of you have been waiting for; the winner of the $50 Amazon gift card. The winner is……Brad Becker! Congratulations to Brad, and again, thank you to everyone who participated in the survey!

Comments 1

Leave a Reply to Brad Becker Cancel reply

Your email address will not be published. Required fields are marked *

*