Lecture 6: Network Attached Storage

Design

• Transparency
• Consistency
• Security
• Reliability
• State and stateful vs. stateless designs
• Service models: download/upload, remote access
• Access semantics: sequential vs. session
• Remote file service, directory service, client module
• VFS interaction
• File Caching

NFS

• Original design goals
• Use of UDP, state
• I won’t ask about the mounting protocol
• Directory & file access protocol; role of lookup RPC
• Approach to caching, validation
• Read-ahead
• Consistency problems
• Open with append & locking problems (original design)
• Lock manager addition

AFS

• Goal
• Core design: whole-file serving & long-term caching
• Cells, volumes
• Namespace and Volume Location Database
• Callback promise & callbacks
• File access semantics
• Referrals

DFS

• Tokens (concept only; you don’t need to know all the types)
• Token revocation

Coda

• Goals
• Volume Storage Group
• Accessible Volume Storage Group
• Resolution
• Disconnected operation mode
• Client Modification Log
• Reintegration
• Hoard database

SMB

• Design goals
• Message block
• You don’t have to know the procedures - just the general idea of what they do
• Oplocks - purpose, revocation, change
• You don’t need to know the batch and filter oplocks
• Read-cache, write-cache, handle-cache leases
• DFS Namespaces

SMB Improvements

• Pipelining (understand the concept)
• Compounding (understand the concept)
• Credit-based flow control

NFS4 Improvements

• State, open
• Compound RPC (understand the concept)
• Locking
• Cache control (understand the concept)
• Callbacks

GFS

• Parallel file systems

• Master - role

• Chunkserver - role

• Replication

• Operation log

• Access model, client interaction Writing

  • chunk lease
  • primary replica
  • Two phases of writing
  • Pipelining
  • Commit

• I will not ask about HDFS

  • You don’t need to know the differences between GFS and HDFS or the names of any of the Hadoop file system components

• I will not ask about Dropbox