Field notes and benchmarks on Linux networking internals, congestion control across high-BDP WAN paths, distributed consensus protocols, and modern NVMe storage engines.
An empirical look at how fq pacing, tcp_notsent_lowat, and disabling slow-start-after-idle eliminate burst-induced tail latency over 150ms+ RTT paths without sacrificing throughput.
Bypassing the full sk_buff netfilter traversal when chaining local TCP proxies on Linux 6.12+ using BPF_MAP_TYPE_SOCKHASH and splice(2) descriptors.
Analyzing p999 commit spikes in multi-AZ Raft clusters: separating WAL fdatasync stalls from network jitter, batching ReadIndex heartbeats, and lease-based linearizable reads.
Comparing leveled compaction, sub-compaction parallelism, and key-value separation (BlobDB) under mixed read-modify-write workloads on PCIe Gen5 NVMe drives.