Tectonic

Key-value stores power the backbone of modern data infrastructure — from user-facing applications to internal analytics pipelines. Yet evaluating their performance accurately remains a hard, largely unsolved problem. Existing benchmarks like YCSB, KVBench, and db_bench fall short in critical ways.

The Problem with Existing Benchmarks

State-of-the-art key-value benchmarks cannot:

1. Emulate dynamic workloads — real applications shift their access patterns over time (e.g., morning read-heavy, evening write-heavy). Static benchmarks miss this entirely.
2. Generate composite keys — production keys often have structured prefixes (user IDs, tenant IDs, timestamps). Uniform random key generation fails to capture prefix distributions.
3. Control data sortedness — ingested data in real systems often arrives partially sorted. Benchmarks that always generate random keys miss this dimension.

These gaps lead to inaccurate performance evaluations and mask how storage engines actually behave under production conditions.

Introducing Tectonic

Tectonic is a Rust-based, highly configurable, and resource-efficient key-value workload generator designed to close this gap. It models the temporal, structural, and dynamic properties of real-world workloads.

Key Features

• Fine-grained operation control: Configure insert rates, update patterns, merge semantics, point/range queries, and point/range deletes independently — and vary them over time.
• Composite key generation: Pluggable strategies for prefix distributions, enabling simulation of multi-tenant and structured-key workloads.
• Dynamic workload shifts: Define workload phases that change composition and distribution at arbitrary intervals, simulating realistic production traffic evolution.
• Sortedness control: Generate data with user-specified sortedness levels — from fully random to nearly sorted — to evaluate storage engine behavior across the ingestion spectrum.

Performance

Tectonic achieves:

• 2× higher throughput than YCSB and KVBench
• Up to 84% lower memory footprint during workload generation

This efficiency makes it viable for long-running benchmark suites without the generator itself becoming a bottleneck.

Impact

Tectonic enables researchers and engineers to evaluate key-value stores under conditions that genuinely reflect production demands — exposing performance characteristics that synthetic benchmarks miss entirely.

Resources

• 📄 View the Paper
• 🐙 GitHub Repository