At Taboola, as in many other companies, it’s all about service availability and latency.
We have almost 700 frontend servers in 7 data centers serving over 500k HTTP request per second. They all expect to get an answer as fast as possible to meet our SLA. The load is not spread evenly throughout the day. The difference between low and high traffic varies up to 10X per geo and per data center.
Moreover, we have ~250 developers that change the code on a daily basis.
Breaking our SLA means Taboola is not serving recommendations - that’s obviously something we can’t afford.
We used a request tracing to find the bottlenecks in our system and limit them using in-house development of TimeKeeper. The TimeKeeper is managing the time budget each component has depends on the load of the server and our SLA limitations.
TimeKeeper – control the latency under extreme load
Full Talk (40 Minutes)
Food & Swag Sponsors
Learn more about each of our Event Sponsors.