The web frontend for VersionEye is a Ruby on Rails application which runs on Puma. For long time it was running on a couple EC2 m3.medium instances, with 2 CPUs and 4 GB RAM and SSD HD. The Puma server was starting 4 worker processes and each of them could have max 16 Threads.
At Christmas time the traffic usually goes down because everybody is spending time with family. That’s why started a small experiment. I downgraded the EC2 instances from m3.medium to t2.small. They have only 1 CPU and 2 GB RAM, no SSD and network performance is classified as “Low to Moderate”.
VersionEye was running on this instances for a couple days and performance wise it was really an issue. The application was really much slower and in the Google Webmaster Tools the crawling errrors increased dramatically, because of timeouts. The network performance was sometimes so bad that I was not even able to SSH into the machines.
So I upgraded the instances again to m3.medium. Now the performance is back to normal 🙂