Tuesday, January 18, 2011

The Riddles and Numbers In Networking

Just when I thought I had seen it all in the world of marketing math I recently witnessed another new low... Scary thought isn't it - I mean in the world of mis-representing the performance capabilities of a product the networking industry has seen some real gems. Maybe I should recap some of my favorites, a few of which I may have even contributed to before becoming a reformed statistician. (Sound's rather Calvinist doesn't it?)

Even as much as I would just love to make liberal sport out of the ne'er do wells I think this would be much more fun as a guessing game. In the 'Comments Section' below please post your guesses. List Company Name, and then the name of the product/system that was positioned this way. If you use a real email address when posting I will send the winner a few T-Shirts and such just for sport.

Also see: Quiz: Separate cyber security fact from fiction

So, to spice it up even more I will throw in an extra few t-shirts for the person who adds one to the list below that is just plain funny or shockingly malevolent.

1) Speciously, this product line claimed a 54Gb/s switching performance - this was the aggregate of a bus, double-counted local switching fabrics, and a 10Gb Cell Switching backplane that was also double-counted.

2) In recent history a form of mathematics has become erroneously known as 'Cisco Math'- however, it started several years earlier, and was NOT in fact first used by Cisco. This is the practice of counting backplane capacity as 2x the actual throughput of a link, usually done so companies can 'fluff' up their performance number. This is the first company and product to do this for an Ethernet switch.

3) Capabilities like local-switching of Ethernet in a modular linecard were awesome innovations in the 1990s, this one is no exception. Equally awesome is the challenge of figuring out how to count what its performance really would be- so with this introduction the vendor felt they should count the backplane capacity, multiply it by 2 to get a 'full duplex number' but then count each local switching module (since it could conceivably forward packets from port to port without traversing the backplane,) and then double that number for posterity sake.

4) Advancing the performance of the Internet this product was introduced with a pretty awe-inspiring distributed chassis system with a fabric interconnect that reached up to 92Tb, although in actual deployments rarely exceeded eight systems interconnected.

5) Lightly sidestepping reality this company recently introduced a module for their switch with 10w/port power draw for a 10GbE interface. This sounds great until you realize this measurement must have occurred without Power Supplies, Chassis, Supervisor, Fabrics, Fan Trays, or even a L3 forwarding engine. This would be like claiming the engine on a Datsun is 1000hp when it is not running, and isn't in a car so its impossible to prove!

6) Ever stretching for larger and more magnificent numbers these guys claimed over 1 Million MAC Addresses in this Ethernet switch, back when most people maybe had 64k support. Turns out they counted the aggregate of the CAM tables on each line card, but if you ever deployed it this way you would have an unsynchronized CAM table and flood traffic all over the place... brilliant....

7) At last some sanity - To counter all of these abilities to misrepresent something that should be grounded in the principals of mathematics and engineering two enterprising souls created an RFC in the IETF to accurately measure and baseline performance. Who are they and which RFC is it?

source

No comments: