[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Rethinking autoconfig, was Re: prefix length determination for DHCPv6
On Mon, 17 Sep 2007 14:21:05 +0200
Iljitsch van Beijnum <iljitsch@muada.com> wrote:
> On 17-sep-2007, at 14:03, Mark Smith wrote:
>
> > What is the motive for reducing the number of packets, and
> > multicast in
> > particular?
>
> Packets use up bandwidth and other resources. The main issue with
> multicast is with wifi and similar networks. Unicast packets are
> acknowledged on those networks and retransmitted at lower speed if
> necessary. With multicast this isn't possible, so those are always
> sent at a low speed and never retransmitted. So they use up more link
> capacity but are less reliable.
>
> In a network such as an IETF meeting network, one or two extra
> multicast packets in the autoconfiguration process can lead to enough
> extra traffic to noticeably impact performance. (Although multicast
> DNS traffic is probably much worse than autoconfig traffic with
> current implementations.)
So what's wrong with using conventional or existing methods to address
or avoid this problem, such as reducing the layer 2 network size,
therefore reducing the number of nodes that can generate multicasts, or
completely abandoning multicast/broadcast operation, by using NBMA
models?
Redoing DHCP seems to me to be a lot of effort when I'd think adequate
and well known solutions to these sorts of problems already exist.
Having a look at the IPv6 NBMA RFC (RFC2491), connectionless
NMBA technologies, such as SMDS, are accommodated. Based on that, I'd
be guessing that wireless ethernet wouldn't be too hard to operate in
an NBMA model if necessary.
Regards,
Mark.