# PagedAttention (vLLM): Block Table + Copy-on-Write + KV-Cache Fragmentation

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-paged-attention-vllm-internals
> Updated: 2026-08-18T21:25:01.921Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part XIII — Custom Kernels & Performance Surgery
**TLDR:** vLLM'in killer feature'i PagedAttention'ın derinlemesine anatomi: KV-cache'i 16-token block'lara böl, logical→physical block table mapping, copy-on-write (prefix sharing), fragmentation %0. CUDA implementation snippets, vLLM source reading. RTX 4090'da prefix cache hit-rate %50+ → throughput +%60.