Getting Hit By The Variable Performance Of The Public Cloud
![VIRTUALIZATION](/assets/categories/virtualization.webp)
In doing many benchmarks of various cloud providers, one of the nuisances is getting slotted on an older/different system of the same instance type. It's been a semi-common issue in my Digital Ocean and Amazon EC2 benchmarking of setting a new instance/droplet on a given tier/type and only to find the performance to be very different based upon the underlying processor. With this latest Digital Ocean testing, I was running the Ubuntu LTS comparison of the $0.476 hourly tier with 12 CPUs, 32GB RAM, and 320GB SSD. One of the systems was powered by CPUs running at 2.0GHz while the other droplet had 2.4GHz CPUs (unfortunately, Digital Ocean conceals the underlying CPU model and just exposes the "QEMU Virtual" identifier). While they cost the same, for performance-sensitive workloads simply your luck can yield very different performance on the same tier. An Amazon engineer previously admitted to me that it's not uncommon to find performance varying by ~10% within a given instance type based upon the age of the hardware. Amazon measures the performance with their EC2 Compute Units, but still that's not too meaningful.
Embedded below and for more data by going to 1404182-KH-VMDIFFS2354 are the results of having a Digital Ocean 12 CPUs / 32GB RAM / 320GB droplet but ending up with very different performance levels. Based upon getting annoyed by the differences and having to create/destroy instances sometimes multiple times across different cloud providers to get matching processors, I've been meaning to do a more extensive article covering the variable performance of the public cloud.
1 Comment