Great! Did you see the post by Howard? He summed up interVLAN routing very
technically in-depth. -Sean
Sean Odom, CCNP, MCSE, CNX-EtherII
Coriolis/Sybex Author
www.TheQuestForCetification.com
E-mail: [EMAIL PROTECTED]
One of the things that confuses this discussion is that "switch" is
more a marketing than a technical term. Certainly "layer 3 switch"
is a marketing term.
There seems to be an assumption in this discussion that fastest is
always best. No. Taking off my Cisco stockholder hat, cheapest that
will do the job is best.
Regardless of the vendor, routing has two distinct functions. Path
determination builds the "routing table," or, more properly, the
Routing Information Base (RIB). The RIB is what you see when you do a
"show IP route." RIBs are optimized for updating by dynamic routing.
Incidentally, the OSPF database, BGP Adj-RIB, etc., are not part of
the RIB, but are inputs to it.
Typically, the first packet to a destination must go through the RIB
to get the FIB set up.
From the RIB is derived the Forwarding Information Base (FIB), which
the second function, packet forwarding, examines to select the
outgoing interface to which the packet is to be sent, based minimally
on destination address.
In process switching, the RIB and FIB are the same data structure.
There is no true FIB.
In fast switching, there is a FIB, which still is in main RAM, and
forwarding is done by the CPU.
In autonomous and silicon switching on the AGS+ and 7000, the FIB was
in a separate memory, and the bus controller (AGS) or Silicon Switch
Processor (7000) did the forwarding. The FIB was on the same board as
the forwarding engine. FIB memory was small, so if the particular
destination was not present (i.e., new or not recently used), there
could be "cache misses". On a cache miss, the FIB was invalidated and
rebuilt from the RIB.
In optimum switching, the FIB and RIB are both on the RSP card, but
in separate physical memories. One processor/memory set does path
determination, and one does forwarding.
In distributed switching (CEF and NetFlow) on router platforms,
simplifying slightly, there is one RIB but multiple copies of the FIB
are distributed onto the VIPs, each of which runs a separate
forwarding process. VIPs have large memories, so the FIB and RIB (at
least in CEF) are in 1:1 correspondence, and there are no cache
misses. Still, the first packet to a destination goes through the
RIB.
In distributed/layer3 switching on "switch" platforms, there remains
a single route determination engine. This can be in the same physical
chassis (e.g., RSM in a 5000), or in a separate chassis (the
"external router"). A Cisco proprietary protocol transfers the FIB
information to a NFFC on a 5000 series or to a forwarding board on
the higher-speed distributed switches.
Using an external router platform as the source of the FIB, or even
using an external router for all inter-VLAN routing, is simply a
design choice. A very real-world situation is having your clients in
one place and servers in another, but on the same VLAN (or using
VLAN-aware NICs). In such a situation, the actual requirement for
inter-VLAN forwarding may be limited to management (e.g., pinging
from the management station) or perhaps email. The function of
routing is important, but not the speed.
In this case, to keep costs low, I'd consider, in order,
1. Use a 2600 to do all inter-VLAN routing
2. Use a 3600 to do it a little faster
3. I don't know the most recent support for external path
determination --
used to be that the 4500/4700 was the lowest platform. Probably a
3600
can these days. Use a 3600 as path determination engine and an
NFFC
or equivalent on the switch platform(s).
4. Use an RSM/NFFC or equivalent.
5. Use a 10000/12000, etc., for very heavy routing loads with multiple
WAN interfaces. A 7200 or 7500 might be appropriate in some cases
-----Original Message-----
From: Chuck Larrieu [mailto:[EMAIL PROTECTED]]
Sent: August 14, 2000 9:56 PM
To: Cisco Mail List
Subject: FW: Switches !!!
Hey, Sean, I gotta say, it has been a while since a thread / discussion has
really struck a chord of excitement in me. A couple of us have also been
chatting off line about this topic. I think those I have spoken to privately
also agree that it is great when there is a topic that inspires one to dig a
bit, do a little research, go back and forth in attempting to understand an
issue or a point or a process, and walk away a little bit smarter. This has
been one fun thread for me, at least.
Thanks, everyone.
Chuck
-----Original Message-----
From: Odom, Sean/SAC [mailto:[EMAIL PROTECTED]]
Sent: Monday, August 14, 2000 7:21 AM
To: 'Chuck Larrieu'; '[EMAIL PROTECTED]'
Subject: RE: Switches !!!
I had it backwards. To route between WAN no to route between VLANs yes.
Sorry it was late!
-----Original Message-----
From: Chuck Larrieu [mailto:[EMAIL PROTECTED]]
Sent: August 13, 2000 11:08 PM
To: Odom, Sean/SAC; 'Jeffrey Humphreys '; ''Frank Wells' ';
[EMAIL PROTECTED]
Subject: RE: Switches !!!
Is it possible that Cisco's Layer 3 switching has evolved beyond the way
things are done of the 5xxx platform. For example, my reading of the product
description of the Catalyst 4908G-L3 switch provides this info:
---------------------------------
The Catalyst 4908G-L3 provides a complete IP routing solution without
sacrificing any of the services that are required to build a scalable
network. The Catalyst 4908G-L3 is a feature-rich switch with full Cisco IOS
implementation that allows network managers to continue to administer and
manage their networks as they do today while scaling their backbone
bandwidths to gigabit speeds. The Catalyst 4908G-L3 supports all the routing
protocols that are used today in mid-sized networks. These protocols
include:
* Interior Gateway Routing Protocol (IGRP)
* Enhanced IGRP (EIGRP)
* Open Shortest Path First (OSPF)
* Routing Information Protocol (RIP) Versions 1 and 2
* Static routes
* Route redistribution
------------------------------------
Now if this guy is an OSPF router, and therefore contains a full table of
the network topology, why does it have to consult an external router to
forward a packet? Doesn't it have its own forwarding table?
I remain unenlightened, and appreciate clarification.
Chuck
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of
Odom, Sean/SAC
Sent: Sunday, August 13, 2000 10:35 PM
To: 'Jeffrey Humphreys '; ''Frank Wells' '; '[EMAIL PROTECTED] '
Subject: RE: Switches !!!
To route between a WAN yes, to resolve VLANs no.
-----Original Message-----
From: Jeffrey Humphreys
To: Odom, Sean/SAC; 'Frank Wells'; [EMAIL PROTECTED]
Sent: 8/12/00 10:53 PM
Subject: Re: Switches !!!
Sean,
I'm confused. Are you saying that if I have a Catalyst 5500 with a RSP
that I will need an additional router (external to the 5500) to route
between VLANs. If that's what your saying, I would have to disagree. I
could do some additional research on it, but I want to ensure that is
what
you are saying.
I believe that the RSP is really just a 7500 and we are running a full
blown
IOS on it to boot. When I do a sho ip ro, I am seeing the local routing
table.
Thanks,
Jeff Humphreys
----- Original Message -----
From: Odom, Sean/SAC <[EMAIL PROTECTED]>
To: 'Frank Wells' <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
Sent: Friday, August 11, 2000 12:03 PM
Subject: RE: Switches !!!
> Fred
> Switches even with an internal route processor cannot take the place
of a
> router. The first packet in Multilayer switching is resolved by the
> external router. The internal route processor learns from the
forwarding
___________________________________
UPDATED Posting Guidelines: http://www.groupstudy.com/list/guide.html
FAQ, list archives, and subscription info: http://www.groupstudy.com
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
___________________________________
UPDATED Posting Guidelines: http://www.groupstudy.com/list/guide.html
FAQ, list archives, and subscription info: http://www.groupstudy.com
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
