Posts

Stylized illustration of stacked sorted runs merging into a single sorted level.

Inside RocksDB LSM-Tree Compaction: Strategies, Trade-offs, and Production Tuning

How RocksDB picks compaction strategies, what each one costs you in write amplification and read latency, and the knobs that matter in production.

Diagram of an LSM-tree with bloom filters guarding each SSTable.

Implementing Bloom Filters in LSM-Trees: Reducing Read Amplification

A practical deep dive into how bloom filters are integrated into LSM-tree storage engines to slash read amplification, with concrete examples from RocksDB, Cassandra, and ScyllaDB.

Diagram of a log-structured merge tree with memtable, immutable memtables, and SSTable levels on disk.

Architecting Log-Structured Merge Trees for Write-Intensive Distributed Databases

How modern distributed databases turn the write-amplification problem into a throughput advantage, and what it costs you on the read path.

Stack of sorted runs in an LSM-tree being merged by a compactor.

Optimizing RocksDB LSM-Tree Compaction: Strategies for Write Amplification and Storage Efficiency

How RocksDB’s compaction strategies trade off write amplification, read amplification, and space amplification — and how to tune them for real workloads.

A compact circuit board with a glowing language model glyph, representing local edge AI inference.

Optimizing Small Language Models for Local Edge Inference: Quantization, Pruning, and Runtime Tuning

A field guide to running small language models locally on CPUs, NPUs, and modest GPUs. Covers INT4/INT8 quantization, structured and unstructured pruning, KV-cache tuning, batching, and how to measure what actually matters.