SCIENTIFIC-LINUX-USERS Archives

March 2009

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Jan Kundrát <[log in to unmask]>
Reply To:
Jan Kundrát <[log in to unmask]>
Date:
Thu, 5 Mar 2009 18:52:41 +0100
Content-Type:
multipart/signed
Parts/Attachments:
text/plain (784 bytes) , smime.p7s (3516 bytes)
P. Larry Nelson wrote:
>  From what I've gleaned about the two protocols from googling, it appears
> that TCP has advantages on a lossy network but that's not our scenario.
> It also is not a stateless protocol, like UDP, so if a server crashes in
> the middle of a packet transmission, the client will hang and filesystems
> will need to be unmounted and remounted.  So it would seem UDP is better,
> at least in our case.

See the famous Why NFS Sucks [1] paper, section 4, to see why you should 
switch to TCP. In short, silent data corruption can happen pretty easily 
with NFS over UDP.

Being able to kill processes waiting on a stuck NFS file handle has 
nothing to do with TCP, but with "intr" mount option.

Cheers,
-jkt

[1] http://www.cc.gatech.edu/classes/AY2007/cs4210_fall/papers/nfsOLS.pdf


ATOM RSS1 RSS2