Skip to main content

IMVU Moves its Metaverse from Redis to ScyllaDB

Hear from Kyle Rudy of IMVU on his team’s success in migrating from Redis to ScyllaDB. 


At IMVU, we had a stack that primarily consists of memcached in front of MySQL and Redis. One of the solutions we had was Redis-based and not scaling terribly well. We set out to replace it with something that would be easier to configure, easier to extend, and , if successful, easier to scale.

We decided on using ScyllaDB at the start of 2017 and found it very easy to install. We got it up and running in production in about a couple of months from zero to full scale  out there ready to go. We immediately saw results that actually weren't quite what we expected. There were some problems, small problems, with the disk usage and the latency that we identified as problems with our schema. We hadn't quite gotten it right the first time around.

That's why we came to ScyllaDB for help and their enterprise package. The first meeting was extremely productive. Within 30 minutes of talking to one of their engineers, we had correctly identified all of the problems that we need to fix. And even before we managed to sign the contract, we put these fixes in place and saw ScyllaDB just beat the snot out of every other competing data source we have tried.

For a prototype, Redis was just fine for the features that we wanted to give our users. But once we actually rolled out to 200,000 concurrent users, the expenses started getting hard to justify. ScyllaDB allowed us to maintain that same responsiveness for a scale 20, 100 times as large as what Redis could handle. We went from an installation of Redis that had a data set of about 350 gigabytes to a ScyllaDB data set that quickly grew twice as large with increased usage, thanks to the simple fact that ScyllaDB is optimized for keeping the data you need in memory and everything else in disk.

Fortunately, ScyllaDB was easy to deploy. Even a guy like me got the project up and running very easily. What issues I did have I managed to get help with even before we signed the enterprise contract. By email they were responsive within a day or two and by Slack, they're responsive within minutes usually.