From: "Norman W. Finn" Subject: 802.1p/D3 ballot comments To: jlarson@fnal.gov, p8021@hepnrc.hep.net Date: Mon, 10 Jun 1996 20:49:11 -0700 (PDT) SUBJECT: P802.1p/D3 - Traffic Class and Dynamic Multicast Filtering Services in Bridged LANs _ X _ I disapprove for the following reasons. __ Norman W. Finn _________ ---------------------------------------------------------------------- I. Incompleteness. (editorial and technical) Section 7 and annex A are both incomplete. II. User_priority based on unicast destination MAC address. (technical) A section E.3.3, discussing the use of 802.1p by higher layers for unicast traffic, would be very useful. Were it present, it would presumably point out that there exists no L3 protocol that is able to map a single L3 destination address to more than one (L2) MAC address. Therefore, if one endstation is to receive unicast traffic at different levels of priority, within the limits specified by 802.1p/D3, its multiple MAC addresses must be algorithmically deriveable from one another. In other words, given the single destination MAC address obtainable by existing L3->L2 address mapping means, a transmitting station must be able to calculate the MAC addresses corresponding to the classes of service to be used when sending to that destination. 802.1p/D3 supplies no specific means to make this calculation. Without a standard algorithm relating an endstation's MAC addresses, interoperability is not possible between endstations. Without using explicit registration of unicast MAC addresses, not even bridges will be interoperable (in the absence of a standard address/priority algorithm). And the penalty for explicit unicast address registration is excessive. As discussed at the Boston interim (6/96) in the contribution, "Applying IEEE 802.1P", the only means available to achieve even de facto interoperability within 802.1p/D3 would be to convert the destination MAC address's U/L bit to a priority indication. 802.1p/D3 does not even hint at this method, and I do not believe it should. The U/L bit has a well-defined meaning in current use. The issue of interoperable unicast class of service indication for 802.3 media must be clarified before I could vote, "Yes". III. Multicast priority/class of service. (technical) Including user_priority in the GARP protocol for multicast, rather than for unicast, MAC addresses, is not fundamentally incompatible with existing L3 protocols. However, if we assume that some acceptable means for expressing user_priority can be found for unicast MAC addresses, that same method should work for multicast MAC addresses. Conversely, since the current means in 802.1p/D3 is unsuitable for unicast MAC addresses, it should not be used for multicast addresses, either. IV. Suggested clarification. (technical) Until the questions raised at the Boston interim are settled, 802.1p/D3 cannot be fixed by a few editing changes. Unfortunately, this means that I cannot suggest any replacement text at this time. Were all references to user priority removed from the document, I would vote "yes" in order to advance the GARP protocol for multicast registration. -- Norm Finn