At 9:11 AM +0000 2/27/03, Steve Wilson wrote: >Thanks for the definition assistance. >The problem with trying to assist in a forum such as this is that if you try >to simplify an answer you end up with an answer that is too simple. All I >was trying to get across was my way of looking at the difference between a >physically separate routing device working at layer3 between subnets and a >chassis like a 6509 which can have individual blades perform the functions >without the limitation of wires between the physical interfaces. The "wires" >are still there, they are just created in software in the chassis. > >Cheers, >Steve Wilson >Network Engineer
Definite cheers in return. As you point out, it's easy to oversimplify--one of the reasons I avoid the L3 switching terminology. It's routing--there are just a wide range of router product design techniques, with different optimizations. Some of these optimizations are for maximum forwarding speed, others are for cost, others for QoS power, etc. To me, a virtual router implies there are at least two distinct control-and-forwarding entities (1 RIB and 1 FIB minimum each). I think this started with a fairly simple CCNA/CCNP level question--I've lost track at this point. My strong opinion is that for the purpose of initial understanding, trying to deal with L3 switching as a significantly different technique is totally confusing for people learning the basics. For them, L3 switching = routing, and then certain speeds, feeds, and feature sets. Choosing those speeds/feeds/feature sets is really a higher skill set for specific network design. The 6500 architecture (as is, for that matter, the 7500, 10000, 12000...) involves a single active control instance and one or more physical forwarding instances in the same chassis, which is slightly different than either of my VR definitions. MLS gets even more confusing when one realizes that on a 5500, the control engine running the routing protocols can be on a completely external box, but the FIB is in a L3 forwarding board in the chassis with multiple L2 blades. 7500s and up actually can have multiple physical L3 forwarding instances. Without even getting into the researchy area of active multiprocessing in the control plane, things are very blurred on how to consider the 6509 by your definition above. I tend to think about it as a set of blades interconnected by a network, which just happens to be in the fabric. In other words, there isn't a huge difference between wires between physical interfaces and fabric, as long as the "wires" are fast enough. There are off-the-shelf optoelectronic fabric chipsets that run at OC-48 and OC-192, but aren't limited to 10 Gbps because they aren't limited to parallel interfacing. In other words, we get very blurred about whether the "box" is still more or less monolithic, or really should be considered an enclosure for a small Storage Area Network interconnecting the blades. I rather prefer the latter. Howard > >-----Original Message----- >From: Howard C. Berkowitz [mailto:[EMAIL PROTECTED] >Sent: 26 February 2003 18:27 >To: [EMAIL PROTECTED] >Subject: RE: L3 Switching Huh???? [7:63728] > >At 3:18 PM +0000 2/26/03, Steve Wilson wrote: >>Charles, >>The 6509 switch needs some configuration in the background to create a >>"virtual router". > > >A bit of a heads-up on this term. It's conceptually useful, but be >aware that "virtual router" was considered to be an alternate VPN >model to RFC 2547, generally promoted by Nortel and Lucent. > >There have been LOTS of IETF arguments about the term. I didn't make >myself popular at one meeting by mentioning "we sure can't define >virtual router, but it's nice we have a virtual router redundancy >protocol (VRRP is the standards track equivalent to HSRP)." > >I was severely corrected that I needed to distinguish between >"virtual router" and "virtual router," depending on whether the >emphasis was on "virtual" or "router." In HSRP/VRRP, the virtual >router refers to a single conceptual router seen by hosts, but is >actually implemented across multiple platforms. > >The VPN people thought of virtual routers as multiple independent >routing (control and forwarding) logical instances on the same >platform. VRF is not quite the same concept, as it assumes more >shared knowledge between routing instances than does a VR VPN. Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=63966&t=63728 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
