nanog mailing list archives

Re: mtu question. more should be better, right?


From: Iljitsch van Beijnum <iljitsch () muada com>
Date: Mon, 7 Nov 2011 18:01:40 +0100

On 7 Nov 2011, at 17:45 , Deric Kwok wrote:

When I setup the server mtu as 9100. why I have to configure the
switch mtu 9300 to make it working?

What this extra 200 bytes is for what purpose? ls it standard?

To avoid problems you really want to set the MTU of all your IP devices on the same subnet to the same value. On the 
switches the MTU needs to be big enough to accommodate the MTU that the hosts and the routers use, but there's no harm 
in it being larger. (Although you may be using up memory for nothing.)

One complication is that not everyone has the same understanding of packet sizes. For IP the MTU is the size of the 
largest IP packet that can be carried, but layer 2 people sometimes add 14 or 18 bytes to that and layer 1 people maybe 
even 38. (14 = ethernet header, 18 = ethernet header + frame check sequence, 38 = header, FCS, preamble, start of frame 
delimiter and interframe gap.)

What is disadvantage of setting our all internal networks (host /
equipment) mtu more than 1500?

Mostly that you'll be hitting path MTU discovery more often. However, this will not cause many issues unlike the case 
where you use an MTU _smaller_ than 1500. Don't set a larger MTU on slow networks (certainly not on 10 Mbps) because 
long packets sit in queues for a long time at low speeds, getting in the way of interactive traffic such as VoIP. Above 
11k the ethernet frame check sequence catches fewer errors.

Current thread: