From: Joel W. Ekis (jekis@xxxxxxxxx)
Date: Thu Feb 24 2000 - 11:00:50 GMT-3
At 08:56 AM 2/23/2000 -0800, Muralidhar Devarasetty wrote:
>Hi all,
>I have small doubt on Multicast MAC address calculation..
>What I understand is the lower order 23 bits of multicast ip address will be
>used with "01-005E-00-00- 00" to generate multicast MAC address.
>But since the multicast ip address range is upto 28 bits of classD I feel
>there will be a clash , I mean same Multicast MAC will be generated for
>different multicast ip addresses.
You win the prize! There is an overlap - 32:1 (that 2 ^ 5). You can thank a c
heap university for the conflict. The developer of multicast, a university pro
fessor (or grad student, I forget) asked for the complete 28 bits, but his boss
wouldn't pay to register all of it. He would only agree to fund 23 bits worth
- this was just an academic exercise after all. (Once again the theory of '64
k of RAM is all you'll ever need' bites the networking industry!!)
Think about TokenRing now. They have ONE 'multicast' MAC address (c0-00-00-04-
00-00)! That's about 269 Million to 1 overlap. Will Win2k run on T/R? Active
Directory uses multicast to function. Interesting problem...
>If this is the case when there will be a packet to one of the multicast ip
>address will it be send to all routers/machines who are having same
>multicast MAC??
PIM won't have this problem, it functions at L3. Flows will be delivered to th
e proper switch. However a well designed NIC should recognize flows that were
delivered to it that are unnecessary and NOT send them to the CPU.
>IF not How will switch differentiate?
The switch can't. The switch only understands MAC addresses and it could be an
y of 32 different flows. If CGMP is used, the router will inform the switch th
at a flow should go to a specific port based on the IGMP L3 information. IGMP
Snooping will allow a non-CGMP switch to read the L3 information in the IGMP pa
cket to decide what flow belongs to a specific port. IGMP Snooping is CPU inte
nsive - the switch must read every multicast packet to see the few IGMP informa
tional packets that are mixed in the flow. New switches like the 6509 that don
't support CGMP use dedicated ASICs to off-load the snooping task from the CPU.
The best book ever written on Multicast is by Beau Williamson - Developing IP M
ulticast Networks. I used this to study for the Lab - it rocks. (Disclaimer -
I'm a Cisco SE, Beau also works for Cisco).
>Any ideas? or my understanding is wrong?
Nope, you are right. Very astute.
>Thanks in advance
>Murali
This archive was generated by hypermail 2.1.4 : Thu Jun 13 2002 - 08:22:54 GMT-3