Large Blob Storage Optimization

1 post · Jun 2026

What this trail covers

ScyllaDB introduces a modified salting technique to address the performance challenges of storing large blobs, aiming to reduce P99 write latency and mitigate head-of-line blocking issues.

All ScyllaDB trails

Jun 2026Using Salting to Lower Latency for Large Blobs in ScyllaDBThis post details a modified salting technique applied to ScyllaDB to improve the performance of storing large blobs. It explains the problem of memory allocation saturation and head-of-line blocking with large cells, presents benchmark results showing significant latency improvements (e.g., 60MB blobs reduced from 1.4s P99 to 63ms P99), and outlines the implementation details of the salting approach. The technique involves splitting large blobs into smaller, manageable chunks and processing them in parallel, effectively treating them as multiple smaller cells rather than one large one. This allows ScyllaDB to leverage its existing efficient handling of smaller data sizes and avoid overwhelming its memory management and connection handling subsystems.
Advertisement