Finding the perfect VM config.

The hot new cryptocurrency trading company “Enfuego Coin” has just taken the plunge into using custom images for their VMs in order to reduce cold-startup time and increase deployment ease. Sadly though, they couldn’t get the right machine type right. Too high, too low, costs all over the place didn’t know how to properly figure this out w/o loads and loads of testing

Although it’s not an amazing performance insight, getting this type of data correct can be the difference between huge successes, and just making the bottom line for a company.

It’s a really common problem for novice, and even veteran cloud companies: wasting resources due to improper vCPU allocation, memory allocation, and incorrect template use. It’s an easy problem to make; Without the right level of analysis, profiling, and benchmarking, it’s really difficult to figure out that you’re never using more than 10% of your provisioned vCPU.

Likewise, it’s a common scenario to over-provision in order to handle spikes in traffic.

Truth is, hitting the sweet spot is really hard, and doing so looks something like this:

Truth is, that’s a lot of work and understanding in order to just figure these things out. And you’ve got better things to do, which is why I highly recommended that Enfuego Coin listen to the recommendations panel.

After running a VM for a while, you may notice some information pop up in the “recommendation” column of your instance list:

This list does all the heavy lifting for you of analyzing your workload, looking at spikes, and what you’re paying, and lets you know that you could be saving money by choosing a different instance size.

And the best part, clicking on the recommendation will allow you to apply the changes directly, w/o having to back and create the instance from scratch, by hand.

Compute Engine monitors the CPU and memory utilization of running virtual machines and makes recommendations using the last 8 days of data on standard and custom machine types.

Upon analysis, recommendations can pop under a number of scenarios:

Also, important to note, is that Compute Engine will only recommend machine types that are available in the zone where the instance is running. So, if moving to a different configuration would require a switch in some resource (that’s not available in your region) it won’t show that data.

These numbers are based on the average utilization of your instances over 60-second intervals, so they do not capture short CPU usage spikes. Applications with short usage spikes might need to run on a larger machine type than the one recommended by Google, to accommodate these spikes.

For Enfuego Coin, after 8 days, they applied all the suggested recommendations, and saved around 23% in total costs for a month. And they did this without having to write any of the diagnostic code themselves!

Related posts:

This article about debugging chrome cast application errors we get once the sender app (IOS or Android) has requested playback to cast app. It also shows how to log whats going on in the receiver app…

The DAO community allow the community to decide what determines a valuable contribution and where resources should be directed to ensure proper incentives and rewards for various types of…

Today I turn 29 years old. Still relatively young in ultimate lifespan, but enough to think that I’ve learned some things. Every year on my birthday, I do two things. First, I write down all of my…