Introduction%20to%20DFS

1
Introduction to DFS
2
Distributed File Systems

• A file system whose clients, servers and storage
  devices are dispersed among the machines of a
  distributed system
• File system operations have to be carried out
  over the network
• A good DFS should ensure transparency
• Clients should have the look and feel of a
  conventional file system

3
Naming and Transparency

• Mapping between the logical and physical objects
• Location Transparency Name and physical storage
  location have no relationship
• Location independence Name and physical storage
  are independent
• Name need not be changed if physical location is
  changed
• Location independent files are essentially
  logical data containers
• Location transparency hides the association b/w
  names and physical storage

4
Naming Schemes

• Combination of host name and local name
• Local name is a path similar to Unix
• Neither transparent nor independent
• Attaching remote directories to the local
  directory
• Popularized by Suns NFS
• Appears as a coherent directory tree
• Globally unique names
• Truly transparent
• Global naming structure spans all names
• Difficult to achieve due to special files

5
Implementing Naming Schemes

• Transparent naming requires mapping between names
  and their associated locations
• Aggregating files into components for scalability
  and manageability
• Hierarchical directory trees
• Replication and caching
• Maintaining consistency of cached view
• Location independent file identifiers

6
Accessing Remote Files

• Needs network data transfer
• Remote service mechanism
• Remote procedure call
• Caching for improved performance

7
Caching

• Idea is fetch once, use multiple times
• If requested data is not available, get it from
  server
• Store fetched data
• Perform access on local data
• Replace data when cache becomes full
• One master copy at the server, several secondary
  copies at clients
• Granularity File blocks to entire file

8
Cache Location

• Main memory
• Workstations can be diskless
• Faster access
• Technology trends memory accesses becoming faster
• Server caches will be in main memory code
  reusability
• Local disks
• Reliability via persistence
• Hybrid schemes
• Best of both worlds

9
Cache Update Policy

• Policy regarding when the modified data is
  reflected on the master copy
• Can have significant impact on the performance
• Write through policy
• All writes are reflected immediately on the
  master copy
• Blocking
• Delayed writes
• Write on flush
• Periodic writes
• Write on close

10
Ensuring consistency

• Ensuring that data being read is consistent with
  master copy
• Client initiated approach
• Clients validates with server whether its data is
  up-to-date
• Frequency of validation is the main issue
• Check on first access
• Check on every access
• Periodic checking

11
Server Initiated Approaches

• Server records the files each client is accessing
• Detects potential inconsistency and notifies
  clients
• Conflicts occur when at least 2 clients cache and
  one is writing
• Invalidation/Update based mechanisms
• Session semantics
• Consistency enforced upon file closing
• Unix semantics
• Consistency enforced upon write

12
Why or Why not Caching

• Locality of accesses
• Gains in performance and scalability
• Big chunks of data lead to lesser overheads
• Disk accesses can be optimized for larger chunks
  of data
• Consistency maintenance is the cost
• Memory/disk space requirements at clients

13
Stateful vs. Stateless Servers

• Stateful servers maintain information about files
  being accessed by clients
• Clients are given connection ids, which acts as
  index into inode tables
• Performance gains Prefetching file blocks
• Stateless servers maintain no state
• Each request is self-contained
• Reliability is the issue !!!