Databricks

What changes when more of your Postgres working set stays in DRAM instead of falling through to slower cache and storage layers? For fixed-size Lakebase Postgres computes with CU >= 80, we now set shared buffers to 75% of DRAM and disable the local file cache, keeping hot pages in the lowest-latency access path. In production, large endpoints have seen up to 2x throughput, fewer reads from the storage layer and lower latency after the change. See how we’re improving Lakebase Postgres compute caching: https://t.co/YxZ9imNSVT