We just cut the "Metadata Tax" by 99%
Published on March 20, 2026 by Suvan Gowri Shanker
Hi everyone,
We’ve just hit a major technical milestone in the Carnelia core. One of the biggest hurdles for local-first databases is the "Metadata Tax"—the reality that sync history often grows faster than the actual data, eventually bricking mobile devices.
We’ve just published our Performance Optimization Report covering our latest implementation of Epoch-based Merkle Pruning (EMP).
The TL;DR on the numbers:
- Metadata Overhead: Reduced by 99% compared to our competitor's baseline.
- Memory Footprint: Significant optimizations in how we handle Merkle-proofs on resource-constrained devices.
- Sync Latency: Delta-syncs are now significantly leaner, especially in high-concurrency environments.
Why this matters: This isn't just a "speed" update. By optimizing how we prune the DAG, we’re making Carnelia viable for long-lived applications that need to run for years without manual database maintenance. We are proving that you don't need a "God-Server" to have a high-performance, consistent data layer.
Read the full technical breakdown here: https://docs.carnelia.dev/release-notes/performance-optimization-report
How your support helped: This research and the subsequent benchmarking were made possible by our current contributors. Your funding directly supported the CI/CD compute time required to run these stress tests across simulated high-latency networks.
We’re now shifting our focus to the WASM sdk and bringing Carnelia to the web.
Thanks for being part of the DAG.
Best, The Carnelia Core Team
🎉 1