50 - Andrew File System.md

AFS Overview

Background

• Developed at Carnegie Mellon University in the 1980s
• Led by Professor M. Satyanarayanan ("Satya")
• Main goal was scalability - supporting as many clients per server as possible

Key Design Principles

• Whole-file caching on client local disk
• Callbacks from server to client for cache invalidation
• Simplified consistency model

AFS Version 1 (AFSv1)

Key Features

• Whole-file caching on client local disk
• Client-side caching of file contents only (not directories)
• TestAuth protocol to check if cached file is valid

Protocol Messages

• Fetch: Get entire file from server
• Store: Send entire file to server
• TestAuth: Check if cached file is still valid

Limitations

• High path traversal costs on server
• Too many TestAuth messages from clients

AFS Version 2 (AFSv2)

Key Improvements

• Callbacks - server promises to notify client if file changes
• File identifiers (FIDs) instead of full pathnames
• Client-side caching of directories

Protocol Changes

• Fetch now returns file/directory contents and sets up callback
• No need for TestAuth - assume cached copy valid until callback

Cache Consistency

• Updates visible on server when file closed
• Server breaks callbacks on other clients when file updated
• "Last writer wins" for simultaneous updates

Crash Recovery

• Clients must revalidate cache contents after reboot
• Server must notify all clients after crash to invalidate caches

Performance Comparison to NFS

Advantages of AFS

• Better performance for large file re-reads (from local disk cache)
• Simpler consistency model
• Reduced server load for common operations

Disadvantages of AFS

• Worse performance for small reads/writes to large files
• Higher overhead for file overwrites

Other AFS Improvements

• Global namespace
• Better security and access controls
• Improved manageability for administrators