All things related to multicast

[Tom Pusateri <>]

In message 
<>
 you write:
>On Thu, Jan 18, 2001 at 02:47:39PM +1100, Markus Buchhorn wrote:
>> At 21:25 17/01/01 -0600, Tony Rimovsky wrote:
>> >Does anyone have advice on a good rate limit value?  Some earlier mail 
>> >from
> 
>> >Matt indicated that 50k SA's in 5 minutes had caused a problem.
>> [...]
>> >two sources, I generated 510 SA's in about 3 seconds (sending from NCSA 
>> >(AS1224) and visible at ACCESS-DC (AS 5640).  All the SA's made it to 
>> >ACCESS-DC and as far as I can tell, that rate generated no problem for 
>> >abilene.  I am obviously leery of pushing this experiment too far.
>> 
>> Is it really the *rate* of SA's that is a problem? Or the total number of
>> SA's? The issue seemed to be cpuhogging when the routers had 50k SA state
>> to deal with under the pressure of a normal background multicast flow.
>> Would it perhaps be better to limit the size of the total SA state? I
>> wonder what would happen if the 50k SA state has been slowly built up (over
>> days, rather than minutes).
>
>  Unfortunately we have to work with what we're given.  JUNOS (and IOS AFAIK)
>  doesn't support limiting the number of SAs (or other MSDP state) at the
>  present time.  However, we can rate limit packets with that source or
>  destinate port.  
>
>Jon

FYI, JUNOS (all releases) will send up to 200 SA's before giving up
the CPU to other tasks. As far as I know, this has been sufficient
to prevent any cpu problems on JUNOS during this storm. If anyone
has evidence contrary to this, please pass it on.

There isn't currently an upper limit to the number of SA's we will
retain but we will not send the same SA out within 45 seconds from
the last time that particular SA was sent.  50k SA's should not be
a problem though.

It is possible (with a minor code change) to add a cap on the number
of SA's but that doesn't allow local traffic to flow when someone
outside your domain is causing the trouble. It seems that filters
on the SA's you accept might allow you to limp along in a more
controlled way.

Or the filter could only kick in when the upper limit is reached
to allow local traffic to flow in the presence of storms.

Remember, MSDP is only a distance-vector protocol like RIP and it
doesn't scale by definition.

Thanks,
Tom