CS 111 Operating Systems Winter 2012

Lecture 16 Redundancy & Performance of NFS

• Professor: Paul Eggert
• Scribe: Wesley Cheng
• Lecture from 3/12/12

NFS - Network File System
To test performance spec.org SPECsfs2008_nfs.v3
Lets look at Sun ZFS Storage 7320 Appliance (May 2012)

Redundancy

RPC(remote procedure call) is used in NFS Server

issues with RPC

• +hard modularity (different address spaces)
• -messages are delayed
• -messages can be last
• -messages can be corrupted

Also, issues with UDP/TCP

• -network might be down, or slow
• -server might be down, or slow

When a Web browser reads a page:

C. Get /HTTP/1.0
S. web page contents

if t/o occurs how should we deal with failed requests?

2 solutions

1. be slow, don't pipeline, wait for response
2. be fast, pipeline, lie to user about whether it worked, write port

use checksums to detect bad packets
if server does detect a bad packet then:
send a response or ask for retransmit

Different actions if no response:

• try again, keep trying. At-least-once RPC (ok for idempotent ops, i.e. read, write)
• return error to caller. At-most-once RPC (for “dangerous” ops)
• exactly-once RPC