Date: Fri, 15 Dec 95 15:45:11 EST From: Anil Rijsinghani 15-Dec-1995 1529 To: 100271.522@compuserve.com Cc: p8021@nic.hep.net Subject: Re: P802.1p/D1 Some initial comments on the 802.1p multimedia bridging draft: - The transmit queue processing algorithm causes a higher priority queue to always pre-empt lower priority ones. While this may be appropriate for safety-critical applications (eg, the "nuclear installation" example), another mode that is needed is one that guarantees some time or bandwidth to certain queues. This very much parallels the CBR paradigm in multimedia networks, and for example could be used for an audio stream. Without this, it would be no different than the situation today when there is significant multimedia activity. So I am suggesting two modes: one in which pre-empting is always done as per the current proposal (perhaps the top priority queue can always pre-empt), and another in which a guaranteed time slice is sufficient. - As far as the recommendation of two queues per port: a minimum of 3 would seem necessary to accomodate differentiation between at least two multimedia streams, or one safety critical, and "other" traffic. - I don't understand how ports in blocking state participate in the scheme. In other words, how a topology change, due to a failed port or LAN, would kick in, in a desired manner. Is there essentially a separate spanning tree per dynamic group address? - I believe that it was mentioned at the last meeting that the GARP protocol can be used to carry priority information even for unicast addresses. Should be called GUARP? - There is an example in which it is suggested that a high-speed pipe might have only one priority, the lower-speed ones may need two. I don't see how it follows that if the high-speed port can accomodate the sum of traffic from all low-speed ones, then traffic prioritization is not useful. - When the GARP mechanism is used for unicast addresses, there would be an undesireable outcome. Since the classification of priority is done based on only the DA, then *all* traffic going to that particular address would end up getting prioritized, whether data or time-sensitive. Some additional qualification may be needed here. - While the text mentions that the server uses a protocol to dynamically establish group address entries in bridges, the GARP protocol only talks about clients. - The state diagrams are very difficult to follow. Text would be very useful to evaluate this section of the specification. Or if the EGMP proposal is required reading for this draft, then maybe it could be included in some way. Regards, Anil