All things related to multicast

["Kevin C. Almeroth" <>]

On Fri, 16 Apr 2004, Hitoshi Asaeda wrote:

>>> I don't know of anything...  it would be tough to measure...  in my
>>> mind, SSM is defined as edge routers speaking IGMPv3 and hosts speaking
>>> IGMPv3.  Sort of hard to measure unless one has a collection point in
>>> every LAN.
>>
>>How about just collecting the number of multicast channels (via SAP
>>message?) and measuring the amount of the data whose destination

Via SAP?  Ugh!  This would not be a good representation of the number
of SSM channels.

>>addresss are in an SSM range, and comparing with whole multicast data
>>in the network (i.e. Internet). 

Not sure how we would compute this (we could determine it at one
router location but not the whole Internet and then how to average it?


>>Because (*,G) joins for the SSM
>>address range are just ignored by the first hop routers and only an
>>SSM-capable node can successfully join them.
>>And yes, we need to trace routing paths for them too.
>>Am I wrong?
>>
>>> If anyone has any good ideas on how to measure this, I'm up for it.
>>> For example, collect stats from SSM sources on who actually joins,
>>> but this has its weaknesses too.
>>
>>Knowing "who actually joins" is difficult, I know.
>>(Just fyi. We made some small work related to this issue.
>>Fethi Filali, Hitoshi Asaeda, and Walid Dabbous,
>>"Counting the Number of Members in Multicast Communication",
>>Proceedings of NGC 2002, pp.63-70, October 2002, Boston, USA.)
>>
>>Thus it may be interested to collect "the number of senders using an
>>SSM addr range" and "the number of the multicast routing paths on as
>>many routers as possible".

Both are tough to collect...  I don't think SAP is a good count.  And
the second, we could do based on PIM forwarding state but at only a
very limited number of routers.  And then knowing there was PIM state
at a router doesn't say anything about the downstream fanout.

-Kevin