Glauber Costa and I will be giving a meetup at AppNexus in New York on Tuesday, November 25th at 6pm.
In this tech talk, we will cover #workloadconditioning which is an umbrella of back-pressure mechanisms that ScyllaDB uses to guarantee stable performance over time with automatic performance tuning.
ScyllaDB is a new implementation of Apache Cassandra which is wire compatible and 10X faster. ScyllaDB uses any trick in the book to achieve superior performance, from it’s own caching, own malloc, log structured allocator and up to a full tcp/ip stack in userspace. However, all these mechanisms are not enough to make it a real contender.
A distributed database needs to cope with failures, spikes, timeout, OOM and many other issues. The approach taken by ScyllaDB is to automatically tune all aspects of the system and adjust them dynamically as a function of the variable workload changes. Below a link from the summit that gives a glimpse about workload conditioning. The session will cover many more low-level aspects and techniques. Speed is nothing without control. You’re welcome to attend!