[e2e] MTU of a loopback interface?

[Vikram Visweswaraiah]

On Thu, 3 May 2001, Jeffrey A. Six wrote:
>e2e,
>
>Me and a colleague have been having a discussion about why loopback
>interfaces have an MTU.  Is this done for any particular reason?  If the
>data being send/received is never actually put on the wire (or other
>physical layer), why does the link layer provide a link layer MTU?  I fail
>to see a reason for this.  Also, if there is such a MTU, how is the value
>determined?  Is this just an implementation issue for those implementing the
>loopback interface code?  What design criteria would go into this
>determination?
>
>Thanks,
>-j6

Well the loopback is intended to provide an abstraction of a real link (despite
it being not sent on the wire and just back into the host) - hence you should be
able to test your networking protocols on the loopback interface without
interfering with others on the link - the portion of your networking code that
needs to determine the MTU, say to decide if fragmentation is needed, still
needs to make some "getMtu()" call and that must not fail simply because we're
addressing the loopback interface.

Now, reg. what value to choose, I don't know - I would imagine you use
some heuristics and decide on a value. In fact, if I were to really test my new
protocol stack, I would like to be able to tweak the MTU at will on the loopback
to exercise my code and make sure it deals with various link layer MTUs fine.

-- 
:=Vikram