1996-04-12 - Re: Protocols at the Point of a Gun

Header Data

From: Scott Brickner <sjb@universe.digex.net>
To: perry@piermont.com
Message Hash: a40909e3ce0a5175758c75783f517b3477e5612a5145807c16787f74c4eab485
Message ID: <199604112313.TAA23689@universe.digex.net>
Reply To: <199604111740.NAA21264@jekyll.piermont.com>
UTC Datetime: 1996-04-12 21:55:45 UTC
Raw Date: Sat, 13 Apr 1996 05:55:45 +0800

Raw message

From: Scott Brickner <sjb@universe.digex.net>
Date: Sat, 13 Apr 1996 05:55:45 +0800
To: perry@piermont.com
Subject: Re: Protocols at the Point of a Gun
In-Reply-To: <199604111740.NAA21264@jekyll.piermont.com>
Message-ID: <199604112313.TAA23689@universe.digex.net>
MIME-Version: 1.0
Content-Type: text/plain

"Perry E. Metzger" writes:
>Scott Brickner writes:
>> Anyway, you computer creates the IP packet, but then sends it to your
>> ISP's router.  That router *always* makes changes to the packet header
>> because it must decrement the time-to-live field and recompute the
>> header checksum.
>There is a trivial trick for making the decrement TTL/change checksum
>operation very fast, based on noting how a decrement would change the
>checksum. Most very high speed routers attempt to avoid doing ANY
>processing of the packets at all beyond this, and IPv6 has no header
>checksum partially in order to reduce this overhead further. Forcing
>routers to do more work is a Very Very Bad Idea.

As I pointed out in a private note to Perry, it's not the high-speed
routers that have to change the packets.  They typically are between
the sort of ISPs that would get "network common carrier" status, and
could rely on the options added (or not) by the other side.  It's only
when the packet crosses the border from outside the "common carrier"
net to inside that the header needs changed, and that's usually at a
terminal server, not a "very high speed router".