Date: Thu, 13 Dec 2001 10:48:30 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: Dan Nelson <dnelson@allantgroup.com> Cc: Mike Silbersack <silby@silby.com>, Geoff Mohler <gemohler@www.speedtoys.com>, hackers@FreeBSD.ORG Subject: Re: NFS: How to make FreeBSD fall on its face in one easy step Message-ID: <200112131848.fBDImUZ70224@apollo.backplane.com> References: <200112130659.fBD6xZt55360@apollo.backplane.com> <Pine.BSF.4.30.0112130215120.60355-100000@niwun.pair.com> <20011213153035.GB56448@dan.emsphone.com>
next in thread | previous in thread | raw e-mail | index | archive | help
:> And if you hadn't heard, Matt just fixed a couple of bugs in the tcp
:> stack which improves NFS greatly. It sounds like after this round of
:> NFS fixes, the first answer to NFS questions should be: Upgrade to
:> 4.5!
:
:I don't even bother with TCP mounts; my default amd rule says
:proto=udp. Is there any reason to add the overhead of the TCP stack if
:you're not leaving your own ethernet?
:
:You should be able to easily saturate a 100mbit link with FreeBSD 4.*
:machines, and I can do 15-20MB/sec with Netgear GA620 gigabit nics (SMP
:2 x pIII/600).
:
:--
: Dan Nelson
: dnelson@allantgroup.com
I think there's a trade-off. TCP mounts deal with heavy parallel
client loads better then UDP because they do real congestion and
streaming control whereas NFS's UDP implementation fakes it. A UDP
mount will work better for a large parallel load from a single client.
I tend to use both types of mounts but I personally prefer TCP mounts
over UDP because they are more secure and easier to get through a
firewall.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200112131848.fBDImUZ70224>