I got interested in computer systems and programming while studying Medicine at MF (Faculty of Medicine), so I decided to switch to FRI (Faculty of Computer and Information Science) and pursue a slightly different career path. While at FRI, I received the Zois scholarship for outstanding achievements (basically for having a very high average grade), which allowed me to buy my first Siemens laptop running a Linux distro.
After finishing my studies, I landed a job at a startup in the telecommunications business, where I was exposed to technologies such as SIP, WebRTC, reverse engineering protocols with the help of Wireshark, etc. Fast forward a couple of acquisitions and a decade or more, and I landed at a gaming company, where I could pick my tech stack for the very first time.
I’m going to present findings from four months of performance optimisations where our team was tasked with creating a transaction service that can handle 40_000 transactions per second on a single node. I’ll present our team’s process and the insights we gained along the way. Much of the performance came from general DB optimisations (Postgres), but I’ll focus on simulating traffic and on critical characteristics of standard library functions. Before achieving the desired performance on the transaction server it was necessary to simulate an appropriate amount of traffic with a distribution similar to our current one.
Talk objectives:
We’ll go from setting up a test server with a dummy endpoint (that does nothing but return 200 OK) to serving 250_000 requests per second to that endpoint. Along the way, we’ll cover changes that delivered from 40% better performance to 10x increases.