Hi Walter, I see some points which you currently do not address when eliminating Layer 2.
- In every network structure we have a layer 2 protocol for a few tasks to do. e.g. IP doesn't provide framing, - finding start and end of packet. This is a task of a layer 2 protocol. On Most WAN-links ist done by PPP (e.g. in POS) or other encapsulation (e.g. RFC2684). - I agree that the addressing structure of ethernet is not any longer needed in this length, but it is a huge effort to change it since every NIC is using it today and backward compatibility has to be maintained. Inside the IEEE 802.3 standard it could be reduced to a two byte addressing scheme, but I believe that NICs are not prepared to work in this mode. Maybe in the loooooong term? - Today and in IP networks layer 2 braodcasts are more or less only used for ARP. There is no other need as long the network supports layer-2 multicast. I see today that in campus networks there is a trend to replace the layer2 switches by layer2/3 switches and use OSPF for load balancing on the links between replacing Spanning Tree. This could be the first step towards the scenario you describe. On the other hand, layer2 switching isn't to bad and really easy to handle compared to a 100 router OSPF network with 5000 networks (prefixes). Maybe we need an innovation in routing protocols before that step. Regards Kai Kai Steuernagel Head of Product Management and Technology Pan Dacom Networking AG Robert-Bosch-Str. 32 63303 Dreieich GERMANY Tel.: +49 6103-932 149 Fax: +49 6103-932 400 Mobil: +49 177-6932 166 (voice mail) Email: [EMAIL PROTECTED] Web: http://www.pandacom.de Nets work together. *** Kai Steuernagel [EMAIL PROTECTED] www.steuernagel.net www.churchofip.org *** -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Walter Zimmer Sent: Monday, December 16, 2002 12:12 PM To: [EMAIL PROTECTED] Subject: Food for thought: Shifting layer 2 functionality to layer 3 with IPv6 Hi! Analyzing the trends in internet technology, which might be summarized as 'towards IP everywhere', I arrived at the following statement: "In a longer timeframe, it makes sense in LANs to incoporate all layer 2 functionality into layer 3." Note: - long timeframe - talking of LANs, not core networks - layer 2 will relly be ethernet - layer 3 will really be IPv6 (I think everybody agrees that this won't come for IPv4 :) Since this is the IPv6 list, I think the people with the most valuable contributions are here. First, lets make a list what functionality ethernet layer 2 includes: A1 - medium access (anyone remembers the good old days of CSMA-CD ? :) A2 - enable local 'routing' (switching) in networking components A3 - broadcast service A4 - multicast service A5 - addressing at layer 2 is needed for autoconfig protocols A6 - ARP is included and necessary for layer 3 operation Note that layer 1 is untouched. There is no better and more cost eficient method of transmitting frames in LANs than ethernet right now, and I think also in the future, because it is designed well and lasting. Here are the arguments supporting the above statement: B1 - In the long run, medium access won't be necessary because ethernet is evolving into a tree based architecture even in cost sensitive areas (e.g. at home) (A1). B2 - Routing in a LAN close to the end devices is not significantly more computing-intensive than switching. Therefore, as computing power gets cheaper, it makes sense to build routing-only devices (A2). B3 - Broadcasts are not necessary in IPv6, because we have a far better mechanism (service-specific multicast) already incorporated (A3). B4 - Multicast is also a available in IPv6. B5 - Autoconifg in IPv4 uses IPv6 broadcast protocols (A5). B6 - Address resolution (also NDP) would not be necessary any more (A6). B7 - Office router software will become more simple, because they won't need layer 2 any more. B8 - If the MAC address is included in the IP address anyway, why repeat it in the ethernet header? Better save the bandwidth. B9 - Security is enhanced: ARP cache poisoning is not possible, because the plug-and-play protocol introduced into IPv6 to propagate addresses would be designed with security in mind. Yes, encryption will come, but ARP cache spamming DoS cannot be prevented without modifying ARP. MAC address locking is no real solution since the administrative overhead is to high. The downsides: C1 - It might really be too early to think about this. However, if everybody agrees that it will come, then it might be beneficial to design current RFCs with that in mind. C2 - Simple switches will need redesign to become IPv6 routers. That's the price for B7. C3 - There will be poblems if the transition strategy is poorly designed. Since ethernet chips today don't insist on sending ethernet headers, seamless transition should be possible. C4 - VLANs would be a problem if the flowlabel could not be used for it. I want to collect pros and especially the cons (since a have so few) for this statement, technical and political, so don't hesitate to express them, either via the list or personal. I'll summarize. Also, pointers to other mailing lists or other information resources would be greatly appreciated. Merry christmas, Walter -- Fraunhofer-Einrichtung Systeme der Kommunikationstechnik (ESK) Walter Zimmer Hansastrasse 32 Dipl.-Inf. D-80686 Munich Telefon: +49(0)89-547088-344 [EMAIL PROTECTED] Telefax: +49(0)89-547088-221 -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] -------------------------------------------------------------------- -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
