An in-depth look at implementing LSM-Trie caching in RocksDB, covering memtable hierarchies, tiered compaction strategies, and production-proven techniques for maximizing write throughput.
Explore how RocksDB’s LSM-tree compaction strategy manages write-heavy workloads by buffering mutations in memory and flushing sorted runs to disk, then merging them through configurable compaction paths with distinct amplification tradeoffs.
Sparse matrix multiplication is a bottleneck in many GPU workloads. This guide covers storage formats, kernel-level optimization strategies, and profiling techniques to extract maximum performance from your CUDA kernels.
A working engineer’s guide to squeezing the H100 with custom Triton kernels: tile sizing, swizzle patterns, pipeline stages, and the warp-specialization tricks that make non-vanilla attention fly.
A practical guide to interpreting flame graphs, understanding their sampling limits, and integrating them into modern observability pipelines.