Re: IP: optimize as a router not host

Ulrich Windl (Ulrich.Windl@rz.uni-regensburg.de)
Mon, 18 Mar 1996 10:23:08 +0100


On 18 Mar 96 at 0:55, root wrote:

> > On 16 Mar 96 at 7:13, Herbert Rosmanith wrote:
> > > reading about "IP: optimize as a router not host" says, that this will
> > > turn off checksum calculation. couldn't this be done automatically,
> > > i.e. for any packet that is known to be routed, don't perform checksum
> > > calculus, while any other packet, that directly addresses this host,
> > > will be checksummed (e.g. any address that contains one of our
> > > interface addresses in the dest field).
> > In case of a bad checksum, wouldn't it be better for the router to
> > discard/reject the packet as compared to forward it. In the latter
> > case an other version of that packet will pass the router again (if
> > the logical endpoint of the VC (virtual connection) complained).
>
> No. If the endpoint gets a bad checksum, it can request the packet again
> immediately.

Yes, but the router will send out the bad (received) packet to the
endpoint, the endpoint will complain and return an error through the
router, the other endpoint will have to resend, the router will have
to route it one more time,...

Can't the ROUTER compain about bad packets and return a NAK (or
whatever) to the sender?

>
> However if an intervening router drops the packet, then the endpoint will
> never see it, and has to wait for a timeout before it can request the
> packet again. This is a BAD idea.
>
> Therefore, 'optimize as router not host' makes a lot of sense, especially on
> hosts with lots of interfaces (e.g. ppp-terminal-server)
>
> -Dan