Bypassing page cache in LSM trees
A deep dive into custom io_uring implementation under sustained write pressure, featuring latency histograms and reproducible microbenchmarks across NVMe arrays.
Register allocation in bytecode interpreters
Dissecting stack-to-register transforms to eliminate dispatch overhead, complete with comparative execution traces and instruction-level branch analysis.
Zero-copy UDP packet pipelines
Evaluating XDP socket performance across high-throughput telemetry clusters, benchmarked directly against standard Linux kernel networking stacks.
Documenting custom mechanical split-chassis routing, RP2040 firmware flashing, and continuous power consumption monitoring under active test cycles.
Reliable systems are quietly predictable under heavy load and uncompromisingly transparent in their failure states.
Systems Lab Notebook, entry 104
All benchmark harness files, open-source crates, and teardown companion code are publicly archived with reproducible container environments.