Comcast Reduces P99 Latencies by 95% with ScyllaDB
Philip Zimich from Comcast explains how his team reduced response times by 95% while reducing its database footprint from 962 nodes of Apache Cassandra to just 78 of ScyllaDB.
Hi, I'm Philip Zimich, Comcast's Senior Director of Engineering . My team supports the X1 DVR platform, powering the business logic behind the X1 DVR and everything that goes into the DVR reminder system. I want to give a description of our journey through DVR as a platform.
We started in Cassandra and we gained some great advantages moving Cassandra. However, the last couple of years we've been looking to reduce our P99, P999, and P9999 times to provide more value back to our end customers. What we found was these long tails have really drastically decreased our user experience on those long ends. And it's been a core goal of ours to change that.
We went through a lot of database evaluations: MongoDB, PostgreSQL, obviously Cassandra. When we found ScyllaDB, we really found a winner here.
We've drastically reduced our node count. We've decreased our P99, P999, and P9999 by 95%. And we're moving forward aggressively as we've rolled this out to our users. What we really love about ScyllaDB is along with our performance gains, there is a huge node reduction across our cluster. We're moving from close to 1000 nodes of Cassandra VMs, down to less than 100 ScyllaDB nodes. We can't say enough about the responsiveness of the team, the responsiveness of the product, and the ability to take our existing workloads and almost "lift and shift" them to the platform with such ease.