This document dives into the architecture and functioning of modern hard disk drives including components like platters, tracks, and heads. It explores disk scheduling algorithms such as SSTF and SCAN, IO timing components including seek time and rotational delay, and strategies to optimize disk IO performance.
Haystack is a custom object storage system developed by Facebook to store and serve billions of photos efficiently. It minimizes disk operations by caching metadata in memory and stores multiple photos per file to reduce overhead. The system balances high throughput, fault tolerance, simplicity, and cost efficiency, supporting large-scale photo sharing.
This presentation details file system data structures such as inodes, their metadata, directory organization, path traversal, and operations including open, read, and write. It also covers caching mechanisms like buffer cache and the role of memory mapping in optimizing file access.
This paper presents Google File System, a scalable distributed file system designed for large data-intensive applications on commodity hardware. It discusses assumptions, relaxed consistency models, architecture with a single master and many chunkservers, replication, metadata management, atomic record append, and fault tolerance in GFS.
This lecture explores integration of IO devices into computer systems, explaining hardware interfaces, classical and modern system architectures, techniques like polling, interrupts, DMA, memory-mapped IO, and the role of device drivers in providing abstraction for generic OS operations over diverse hardware.