Talks

What this talk covers

• Building a strongly consistent, distributed Key-Value engine from scratch in Go on top of BadgerDB and Multi-Raft
• Isolating lock contention: Moving from a global mutex to partitioned lock groups using xsync and cache-line hash tables
• Bypassing the two-phase commit (2PC) tax for single-partition writes
• Defensive range queries: Routing chunked, client-driven iteration to eliminate memory buffering and avoid API layer OOMs
• Parallelizing Protobuf serialization and multiplexing persistent connections to drop tail latencies

Key takeaway

What is fast on paper is almost always slow in reality, until we profile, measure, and adapt to the hardware underneath. By identifying coordination bottlenecks and CPU-bound serialization limits, we optimized our distributed key-value engine from a sluggish start to a p99 write latency of ~5.6ms at 10 000 TPS.

What this talk covers

• Common claims databases make about durability and read/write optimization
• How to evaluate those claims critically using real design patterns
• Core storage engine patterns: WAL, fsync, B+tree, bloom filters, LSM trees, WiscKey

Key takeaway

Every database makes claims, “Durable writes”, “Fast reads”, “Horizontal scale”. But do you know what’s actually happening underneath? This talk gives you the vocabulary and the patterns to stop taking those claims on faith and start evaluating them yourself.

Target Audience: Engineers with basic Rust knowledge looking to go deeper into systems programming.

What this workshop covers:

• Memory allocation and ownership in Rust
• Zero-cost abstractions in practice
• Concurrent data structures: design and implementation
• Building a working concurrent cache end-to-end

Key takeaway

Building a tiny concurrent cache is a journey of countless design choices. But when you treat the Standard Library as a mentor and base those decisions on its idioms, you are adopting the collective engineering wisdom of the Rust community.