Transcript A4-Presentation3

Sun Network File System
Presentation 3
Group A4
Sean Hudson, Syeda Taib, Manasi
Kapadia
Problem
How can physically distributed
computers share a file system?
Solution
Sun’s Network File System (NFS)
Background
Developed by Sun.
The NFS specification is platform
independent:
– Portable
– Operates in a heterogeneous environment.
Fundamentals
NFS was developed to allow a machine to
mount a disk partition on a remote machine
as if it were local.
This allows for fast, seamless sharing of
file across network.
There is no global naming hierarchy.
NFS allows any machine to be client or
server.
NFS Protocols
Mount Protocol
File Access Protocol
Their functionality is defined as a set of
Remote Procedure Calls (RPC).
Mount Protocol
Client
home
Server
usr
Request
Dir1
Dir2
Open
Mount
Local
Mount Protocol (cont.)
After mount request, server returns a file
handle (a key for further access).
Mount operation changes only the user’s
view.
NFS Protocol
Client
Server
RPC Calls
• Search
• Read/Write
• Manipulate links/directories
Implementation
Three major layers of NFS architecture:
System Call layer:
– This handles calls like OPEN, READ, and
CLOSE.
Virtual File System (VFS) layer:
– distinguishes local files from remote ones.
NFS service layer:
– implements the RPC calls.
Schematic view of NFS
Client
Server
System call layer
Virtual file system layer
Virtual file system layer
Local OS
NFS client
NFS server
Local OS
Local disk
Message to
server
Message
from client
Local disk
Pathname Translation
This is done by breaking the path into
component names and performing a
separate NFS lookup for every pair of
component name and directory v-node.
Lookups are performed remotely by the
server.
Performance
Clients cache file attributes (i-nodes) and
file data.
Cached data blocks and directories are
discarded periodically.
When a cached file is open, the time it was
last modified is checked at the server. If
necessary, a new copy is brought from the
server.
Performance (cont.)
Servers keep a cache to reduce disk I/O.
Read-ahead caching
Write-behind caching
– Periodically, modified cache blocks are written
back to servers.
Issues
Synchronization of file accesses are not
well-defined.
Widely criticized for not implementing the
UNIX semantics.
Conclusion
NFS is a simple, popular and widely used file system!
References
 Casvant, Thomas and Singhal, Mukesh. Distributed
Computing Systems. IEEE Press.
 Mullender, Sape. Distributed Systems. ACM Press.
 Pradeep, Sinha K. Distributed Operating Systems. IEEE
Computer Society Press.
 Stallings, William. Operating Systems.
 Tanenbaum, Andrew. Distributed Operating Systems.