#ident "@(#)issues	1.5 07/06/06 SAC"

PSARC 2007/272: 	Project Clearview: IPMP Rearchitecture
Submitter:	Peter Memishian
Owner:		Bill Sommerfeld
Exposure:	open

Issues for inception (06/06/2007):

jdc-1	For the preferred new /etc/hostname.ipmp usage, are there
	documentation changes?  How do we encourage users?

jdc-2	Instead of privacy extensions, consider using the group or
	ipmp interface name (and perhaps hostid) as a 'seed' for the
	IPv6 random interface token.  That way, the system will be
	likely to acquire the same random address each time -- a more
	friendly behavior for administrators.

jdc-3	Nit: does the ifconfig "FAILED" flag really appear first on
	the line as shown on p7 of the design?  I thought the flags
	were traditionally in ascending order, and IFF_UP (0x1) is
	thus first.

jdc-4	Nit: contrary to 4.1.11, it'd be possible to make auto-revarp
	work for the group.  It's probably not worthwhile, though.

jdc-5	Nit: IFF_VIRTUAL isn't intended to be ignored by routing
	daemons (p38).  Instead, it means that only the local address
	means anything.  No output hardware is available, so you can't
	forward "through" it, but you can advertise the address as
	/32.

djr-0	This case does not change the underlying architecture of IPMP,
	which is a chunk of code bolted onto/into IP and continues to
	deliver its functionality via virtual interfaces.  Virtual
	interfaces (e.g. lo0) are confusing for customers to deal with
	and lead us to delivering new projects (such as PSARC/2006/475)
	that invent new ways in which we erect a smoke screen and try
	to make them appear to behave "normally."

	The correct rearchitecture of IPMP would be presenting the IPMP
	interfaces as GLDv3 interfaces, allowing them to be managed and
	observed like all other network interfaces using the standard
	set of tools and APIs, without any need for new, "special" ones.

	See 3.3 of PSARC/2005/132 (Clearview Umbrella Inception):
	"This project is targeting a drastic simplification of Solaris
	 administration by providing consistent network and virtual
	 interfaces."

wes-1	section 4.6.2: rfc3041 uses random identifiers in host addresses
	for privacy purposes which intentionally change over time to
	make it harder to do traffic analysis of hosts which move
	around the network.  For use on a server you're going to need
	stable addresses; where is this information kept persistently?

wes-2	what other than tradition prevents use of point-to-point interfaces
	with IPMP?  

wes-3   section 3.19.2: "CGTP must guarantee that each packet sent using a
	particular source address is always sent over a specific interface"
      	I thought CGTP was raid-1 for networks, sending each packet
      	over multiple interfaces and paths. are we talking mac address or IP
      	address here?