Building Ultra-Performant Message Streaming

Piotr Gankiewicz

In high-performance distributed systems, every delay at the infrastructure layer compounds across the many services exchanging messages. Unpredictable tail latencies can seriously degrade performance and user experience.

Requiring hundreds of megabytes of memory just to spin up a virtual machine before your tooling even starts? That's madness. Relying on GC-based languages with unpredictable pause times to build critical infrastructure? That's a shot in the foot. Modern hardware is incredibly capable, but it demands modern solutions.

Let's talk about the architecture built for these needs: thread-per-core, shared-nothing, io_uring, zero-copy, Viewstamped Replication. These are just a few of the building blocks that allowed us to achieve sub-millisecond p99+ latencies in Apache Iggy.

Piotr Gankiewicz is Co-founder & CTO at LaserData | Apache Iggy PPMC