Date: Fri, 1 Dec 1995 16:48:07 -0800
From: hendel@uask4it-223.Eng.Sun.COM ( Ariel Hendel)
To: P8021@nic.hep.net
Subject: 801.p/D1 comments


Here are my comments on the 802.1p/D1 draft distributed by Tony in
Montreal.


802.1p/D1 General Comments:

1) Two types of Service Primitives are described in 2.6.9.x, with their
corresponding database entries in 3.9.3.2 that allow:

	a) recipient registration (upstream)
	b) server registration (downstream)


The mechanisms for recipient membership information dissemination are
covered in Chapter 9 (and in EGMP). I haven't found a description that
covers server registration, and how this information is propagated
downstream as mentioned in 2.6.9.2


2) The association of priorities to Destination Address must also work
for unicast addresses (and it is indeed covered in 3.9.3.1). When
dynamic entries with priority information are aged (for example due to
a topology change) they cannot be "removed" from the database without
losing the priority information. 

While a Multicast Priority aging mechanism exists in 3.9.3.2 (when
server registration has not occured for a while), there is no
equivalent mechanism for aging the Priority information of a unicast
address.


3) Maybe it is me, but I am confused by the Topology Enforcement
Filtering "per input port" description. The Filtering Database is
common to all ports (per 802.1d Fig 3-4), and whenever the Filtering
Database is described (3.9 for example) it is a Destination Address to
output port/priority function.  For example, I interpret "given Port"
in 3.7.2 b) as the output Port of the first paragraph.

However 3.9.1.1 b) says that the filtering decision is also a function
of the inbound Port over which the packet was received.


Details:

4) Is PORT_SET in 2.6.9.3 the same as PORT_MAP in 2.6.8.2 ?  What do
these field include, just output ports or input to output ports
mappings?

5) 3.9.3.2 specifies aging based on elapsed time, while 9.3.2.1
describes garbage collection based on LeaveAll received from some other
bridge.  As I mentioned in Montreal I think the LeaveAll scheme is too
sensitive to packet loss, and could cause oscillations when the GARP
packets go through a congested bridge with basic filtering only.

6) In 9.6.2.8 recipient GARP Join PDUs convey priority, while the
2.6.9.1 and 2.6.9.2 primitives only let the server primitives control
priority.



Ariel Hendel
Sun Microsystems