Subject: Re: ip reassembly time exceeded?
To: Matthias Scheler <email@example.com>
From: John F. Woods <firstname.lastname@example.org>
Date: 01/27/1997 08:36:16
> > I heard that  is the minimum length specified for IP packets.
> No, the minimum length is 40 bytes.
RFC 760 (the IP spec) says:
All hosts must be prepared to accept datagrams of up to 576 octets
(whether they arrive whole or in fragments).
which I think is an implies that refusing to accept a 576 octet whole packet
is in violation. However, the Hosts Requirements document (RFC 1122) only
firmly states that the fragmentation reassembly buffer must be at least 576
octets, and just recommends:
Since nearly all networks in the Internet currently
support an MTU of 576 or greater, we strongly recommend
the use of 576 for datagrams sent to non-local networks.
A router which refuses to fragment below 576 bytes is just plain broken,
however, in my humble opinion. Unless, of course, the originating host is
sending 576 byte IP packets with the dont-fragment flag set (which it is
entitled to expect should get through, but is a pointless thing to do).
Typically the sub-500-byte MTUs are to achieve .2-second packet times per
Van Jacobsen's interactivity recommendation. At 14.4Kbps, that's 320 bytes
(assuming you run at least 19200 to the modem to make up for the framing
bits the modem isn't sending). At 28.8Kbps, that's 640 bytes. Note also
that the interactivity hints only really works if your connection sorts
small traffic ahead of large traffic (otherwise your interactive traffic
just piles up behind multiple bulk transfer packets), and a lot of routers
don't do that (especially "bridging routers", since packet reordering is
officially EVIL in the bridging world).
I usually run 1072 on my ~24Kbps link, and I find that it sucks no worse
than when I have it at ~300 bytes. I do enough bulk traffic that I'd
rather not send the extra headers...