ScyllaDB, the database for predictable performance at scale, today announced ScyllaDB 2024.2, the first enterprise release featuring ScyllaDB’s new “tablets” replication architecture. This new architecture, part of a multiyear project to implement and extend the Raft consensus protocol, delivers unprecedented levels of elasticity and efficiency. It builds upon ScyllaDB’s signature shard-per-core architecture, which is known for providing predictable low latency at scale (e.g., steady single-digit millisecond P99 latency even with workloads exceeding 2M ops/sec). [See tablets in action]

“In the era of AI, databases need to rapidly grow and shrink as millions of users access the applications powered by them,” explained ScyllaDB co-founder and CEO Dor Laor. “Often, teams are forced to overprovision the database to meet the demand. Elasticity speed is measured by how quickly a given deployment can start handling increased throughput and volume in production. With ScyllaDB’s elastic speed, customers can tighten their deployment, use what they really need on a per-minute basis, and be prepared for any given spike in demand.”

“People tend to think that databases are not scalable, but the reality is that they’re not very elastic,” explained David Malinge, Senior Staff Software Engineer at ShareChat. “I’ve seen databases being well overprovisioned to manage tests, backfillings, or even holiday traffic months ahead. With ScyllaDB, we scale our clusters up and down many times in the year. That really helps us manage our costs. And although we don’t use it in production yet, ‘tablets’ is a really solid step towards increasing elasticity.”

Additionally, the enterprise release offers new capabilities designed to help enterprise customers reduce infrastructure costs and streamline operations:

  • File-based streaming for tablets accelerates scaling operations (e.g., adding and removing nodes) up to 30X. 
  • The new design enables scaling at the network line rate – without any dependency on the schema.
  • Improved network compression for intra-node RPC. 
  • Improved storage utilization thanks to the inherent efficiency of tablets. 

“Customers can now live closer to the edge with respect to their storage capacity. Previously, we would recommend that you fill your disks to about 70% and even set the alerts a little earlier,” explained Avi Kivity, ScyllaDB co-founder and CTO. “Now we can have clusters filled to 90% or 95% of the disk capacity.”

These improvements are especially welcome for ScyllaDB’s customer base, which is dominated by fast-growing companies whose competitive advantage depends on instantaneous user experiences at extreme scale.