[pim] Some comments on draft-ietf-pim-assert-packing-00

[Stig Venaas <[e-mail]>]

Hi authors

The draft looks pretty good to me. The security considerations need
some more details though.

I want to follow-up on a comment I made at the meeting.

When sending a packed assert one can include multiple records. In
order to pack as many asserts as possible, it is OK to send the
periodic assert winner messages slightly earlier than usual, but they
should not be sent later.

When the assert timer expires, it is important that an assert is sent,
and received by other routers within 3s (default value of
Assert_Override_Interval). It might be OK to delay sending some
asserts a few ms so that they can be packed with asserts shortly
after, but it cannot be delayed for long.

However, there is no harm in sending an assert a few seconds before
the assert timer normally would expire. E.g., if one assert is
supposed to be sent in 50s and another in 52s, it is OK to send them
both in 50s. One should then reset the assert timer for both of them
so they would expire at the same time later. This could be achieved in
a few different ways, but we don't have to go into the implementation
details.

What I'm suggesting is some brief text explaining that one should not
delay sending asserts in order to achieve optimal packing (maybe can
say that a few ms is ok), but there is no harm in sending some asserts
a few seconds earlier to achieve better packing.

Maybe another way of saying this is that asserts happening within a 1s
interval can be packed together. With a lot of asserts happening,
which is the use case for packed asserts, one may get efficient
packing just by considering 1s intervals. But this should ideally be
done sending asserts up to 1s earlier than usual, rather than up to 1s
later.

Regards,
Stig