Scylla (database)


Scylla is an open-source distributed NoSQL column-oriented data store. It was designed to be compatible with Apache Cassandra while achieving significantly higher throughputs and lower latencies. It supports the same protocols as Cassandra and the same file formats, but is a completely rewritten implementation, using the C++17 language replacing Cassandra's Java, and the Seastar asynchronous programming library replacing threads, shared memory, mapped files, and other classic Linux programming techniques.
Scylla uses a sharded design on each node, meaning that each CPU core handles a different subset of data. Cores do not share data, but rather communicate explicitly when they need to. The Scylla authors claim that this design allows Scylla to achieve much better performance on modern NUMA SMP machines, and to scale very well with the number of cores. They have measured as much as 2 million requests per second on a single machine, and also claim that a Scylla cluster can serve as many requests as a Cassandra cluster 10 times its size - and do so with lower latencies. Independent testing has not always been able to confirm such 10-fold throughput improvements, and sometimes measured smaller speedups, such as 2x. A 2017 benchmark from Samsung observed the 10x speedup on high-end machines - the Samsung benchmark reported that Scylla outperformed Cassandra on a cluster of 24-core machines by a margin of 10–37x depending on the YCSB workload.
Scylla is available as either an on-premises solution, on the major public cloud providers or as a DBaaS.

History

Scylla was started in December 2014 by the startup Cloudius Systems, previously known for having created OSv. Scylla was released as open source in September 2015, under the AGPL license. Employees of ScyllaDB Inc. remain the primary coders behind Scylla, but its development is open to the public and uses public GitHub repositories and public mailing lists.