CFP: RustINDIA Conference 2026

Introduction (5 mins)

• Case study: where we hit Java’s performance wall in our OLAP engine
• Why Rust made sense for rewriting the execution layer

The Comparison Program (5 mins)

• Implementation: HashJoin on primitive types
• How it works: build phase (insert into hash map), probe phase (lookup + materialize)
• Why this is a good benchmark: real-world engine comparisons (e.g. our Java executor vs Rust executor) use different algorithms, data structures, and architectural choices — so performance differences reflect design decisions, not language differences. This benchmark isolates the language by running the exact same algorithm in both, making it a true 1:1 comparison of runtime characteristics

Rust vs Java: Default Implementation (5 mins)

• Why this experiment matters: Rust treats performance as a default goal. Companies building performance-sensitive products benefit from saner defaults that take you far without manual optimization
• Using default hash maps and hash functions from each language’s standard library, no pre-sizing, no manual optimization
• Results and what factors cause the difference

Rust vs Java: Same Implementation (10 mins)

• Both use the same hashing algorithm, open-addressing hash maps, primitive arrays - squeeze more performance out of java
• New results and what factors cause the difference: escape analysis failures, auto-vectorization, JIT overhead, instruction counts, and cache behavior

Conclusion (5 mins)

• Summary of results across all variants
• Rust’s defaults are more performant out of the box
• For performance-sensitive products, saner defaults mean you travel further before needing manual optimization

Speaker bio

Hacking on DB@e6data