case studies
writeups of performance investigations — the hypothesis, the tooling, the finding, and the numbers. all grounded in real production workloads.
-
a 40% json parsing speedup in apache pinot, found hiding in plain sight
how pre-production profiling surfaced a two-line optimization that had been hiding behind three years of "good" refactors. jackson was parsing every message twice.
-
uncovering 3 performance optimizations in apache pinot's protobuf ingestion
a customer flame graph at 2–3M events/sec revealed a rocksdb data-block cache miss, redundant protobuf field lookups, and a reflective parser that had a 2.9× faster alternative.
just out: startree iceberg benchmarking & scale tests — 12k to 125k qps on open tables.
coming soon: the internal real-time database benchmarking framework writeup.
for the short list of existing writing, see writing.