Performance Optimization Techniques: Indexing, Caching, Compression, Prefetching, and More

This article explains that software performance optimization is fundamentally about trading time for space (or vice‑versa) and making informed compromises before a system reaches a bottleneck.

Indexing Techniques : Using extra storage to reduce query time, with data structures like hash tables, binary search trees (red‑black trees), B‑Tree, B+Tree, LSM Tree, Trie, and inverted indexes, and best‑practice tips for primary keys, composite indexes, and avoiding over‑indexing.

Caching Techniques : Adding storage layers (DNS, OS, CDN, application, CPU caches) to exchange space for latency, discussing cache invalidation challenges, cache penetration, thundering herd, cache avalanche, and object pooling as a cache variant.

Compression Techniques : Time‑consuming lossless compression to reduce data size, with use‑cases in HTTP gzip/deflate, HTTP/2 HPACK, JS/CSS minification, binary protocols, and discussion of information entropy, lossy compression, and extreme data reduction (e.g., tree‑shaking, HTTP/2, header minimization).

Prefetching : Anticipating future data accesses to improve perceived latency, illustrated by video buffering, HTTP/2 server push, warm‑up caches, and pre‑allocation of hot resources, while noting side‑effects such as increased startup time.

Load‑Shedding (削峰填谷) : Smoothing traffic spikes by delaying or batching work, using back‑pressure, rate‑limiting, message queues (Kafka, RocketMQ), and avoiding error‑storm cascades.

Batch Processing : Grouping operations to amortize overhead, with examples ranging from JS bundling, requestAnimationFrame, database bulk inserts, Redis MGET/MSET, and system‑level I/O batching, plus guidance on optimal batch sizes.

Parallelism and Scaling : Horizontal scaling (八门遁甲), sharding (奥义), lock‑free techniques (秘术), and hardware‑aware optimizations such as CPU affinity, DMA, zero‑copy, and choosing appropriate instance types.

Operational Insights : Hardware latency numbers, memory layout costs, network overhead, and the importance of measuring bottlenecks before applying optimizations.

Conclusion : Emphasizes ROI‑driven optimization, continuous monitoring, and avoiding premature or excessive tuning, while acknowledging that many performance gains require trade‑offs and careful engineering.