Behind the Massive Scale Required for Content Recommendation - CitC Episode 252
06 Aug 2021
About this Podcast
Taboola's Ariel Pisetzky joins host Jake Smith to talk about using artificial intelligence (AI) for personalized content recommendations at massive scale, reaching up to 4 billion web pages a day. Ariel discusses why Taboola runs their services on their own on-prem infrastructure of over 10,000 servers with Intel processors, many of which are dedicated to inferencing workloads. Ariel also goes into detail about some of the performance optimizations the company has found gen-over-gen with increased memory access. Jake and Ariel close the episode by talking about the open source community and how to respond to a crisis, like a fire breaking out inside your data center.