[c-nsp] Router/Switch Sizing terms

[Justin M. Streiner]

On Mon, 8 Jan 2007, Pete Templin wrote:

>> 3) Switching Capacity
>
> L2 devices switch packets from ingress to egress.  L3 devices route, and
> then switch packets from ingress to egress.  Either method requires an
> appropriate (L2, L3, etc.) lookup, one time per packet regardless of
> packet payload size, and therefore most switches and routers are
> performance-limited by the number of frames/packets they can perform
> switching/routing lookups on (and then switch the packet to the egress
> port).
>
>> 4) Forwarding performance
>
> Often a measure of switching capacity multiplied by minimum, average,
> weighted, and/or maximum packet size.  Used to differentiate between
> devices that can only fill their ports if maximum-sized packets are
> used, and devices that can fill their ports if minimum-sized packets are
> used.

Also, sometimes it's well worth taking the performance numbers that a 
vendor gives you with a healthy grain of salt.  A vendor can make a claim 
like "Our new ABC9000 platform will forward up to 40 million packets per 
second", but without knowing exactly what this number means and how they 
arrived at it, i.e. what tests they used, etc, it's at best disingenuous 
and at worst flat-out wrong.  Packet forwarding tests need to discuss 
forwarding performance at a wide range of packet sizes.  This is 
especially true for a router, or a switch that has routing functionality.
Small packets, especially lots of small packets, are generally more 
taxing, especially on a software-based router.

Also important to know are:
* Does the device (route|forward) in hardware or software?
* If it does so in hardware, what conditions exist that could cause it to 
(route|forward) in software?
* What is the performance impact of doing so?

If you're dealing other features like NAT or IPSEC, those will often have 
their own sets of performance numbers.

jms