NFS Performance Tuning for HP-UX 11.0 and 11i Systems

nfs performance tuning for hp-ux 11.0 and 11i systems page 95

Notes:

Page 95July 22, 2002

nfs/udp

vs.

nfs/tcp

• Protocol-Induced Overhead

• Retransmissions and

Timeouts

• Network Switch Buffering

Considerations

• Should you use NFS/UDP

or NFS/TCP in your

environment?

One of the design goals of the original version of NFS, and every version since,

has been to minimize the amount of network latency involved in accessing remote

files. For this reason NFS was originally designed to run exclusively over the UDP

network transport, as this transport provided a lightweight delivery mechanism.

Over time, as clients and servers became geographically dispersed across wide

area networks, it became necessary to provide guaranteed data delivery and

improved handling of network timeouts and retransmissions – even if the added

overhead meant slower performance in local area networks. The TCP/IP protocol

provided these benefits and was already widely in use by other networking

applications. The developers of NFS therefore decided to modify the NFS protocol

to allow it to run over either TCP or UDP and allow the system administrators to

decide which transport mechanism better met their needs on a per-mount basis.

Now that both UDP and TCP are supported protocols, the question is ‘Which one

should I use?’ To answer this question, a system administrator needs to understand

the differences between the two transport mechanisms and how these differences

can benefit or hinder NFS performance in their environment.