Mark Wong @markwkm.bsky.social · 25/09/2026Testing impact of shared_buffers with 32,000 customer TPC-E database that is about 400GB in size on-disk. Interesting how the i/o spikes and hurts throughput when shared_buffers > 32GB and < 72GB on this 128GB system. 010
Mark Wong @markwkm.bsky.social · 10/09/2026Worked through some DBT-5 kit improvement before getting a feel for the Throughput vs Users at the minimum scale factor. 16 users matches the number of logical processors on this r5b.4xlarge instance type. Continuing to scale up the database size. 000
Mark Wong @markwkm.bsky.social · 06/08/2026Starting to size TPC-E database on r5b.4xlarge instance type on minimum scale factor of 5000 customer. PostgreSQL can serve 148 Trade Results per second suggesting this system can support somewhere between 5000 and 75,000 customers. Narrowing it down... 000
Mark Wong @markwkm.bsky.social · 05/08/2026Sizing up r5b.4xlarge instance type for TPC-E-like workload. Some grossly set PostgreSQL GUCs to start: shared_buffers = 16GB, effective_io_concurrency = 200, max_wal_size = 500GB, random_page_cost = 1.0, checkpoint_timeout = 30min 000
Mark Wong @markwkm.bsky.social · 10/07/2026Containerized DBT-7 (Fair Use TPC Benchmark(TM) DS): docker pull ghcr.io/osdldbt/dbt7:latestghcr.io 010
Mark Wong @markwkm.bsky.social · 27/03/2026pg_plan_advice differences with enable_nestloop off for TPC-DS queries gist.github.com/markwkm/3775...gist.github.comCompare pg_plan_advice DBT-7 / TPC-DS SF 1Compare pg_plan_advice DBT-7 / TPC-DS SF 1. GitHub Gist: instantly share code, notes, and snippets. 010
