I have a few question regarding chassis devices and some more CEF concepts (distributed CEF). I haven’t seen a lesson related to distributed CEF, so I am asking here.
First of all, what exactly is a chassis? I am not a native english speaker and it’s the first time I am hearing this word.
So when you buy a chassis switch or a router, the idea is that you’ll be constantly able to rebuild it over and over based off your needs? Since I’ve heard that they’re really customizable, you can remove/add any modules, interfaces, etc.
A linecard is something that you’d typically insert into these devices and it’s basically a card that provides interfaces on the device?
Processor “cards”? This is the first time I am hearing about this. What is a processor card? I imagine a CPU being something you plug into a CPU socket, I’ve never seen it as a card before. Does its function differ in any way from regular CPUs?
Also, why is a processor card used for packet switching decisions? What about CEF, the ASIC, and the entire idea behind doing forwarding only in hardware?
Again, why is the RP card involved in packet forwarding? I’ve seen entire lessons talking about how important it is to forward data using ASICs, yet we seem to be going back here again.
How can a line card make forwarding decisions? Does the line card contain a chip like an ASIC, or?
That’s all! By the way, I think it would be great if Rene created a lesson for Chassis devices and Distributed CEF as I believe it’s an ENCOR topic and other resources include it!
I will attempt to respond to each of your questions below:
A chassis, in the context of networking, is a metal frame that houses modules like switch, router, and firewall blades. This is a modular design that allows for scalability and flexibility, as you can add or remove modules as needed. For example, the Cisco 9400 switch is a chassis switch. It has slots that you can place various blades into, where each one provides different interfaces and/or capabilities.
You wouldn’t constantly rebuild it over and over (although you can), but you would typically purchase the chassis, and the cards that you need today. Since it is modular, you can expand it over the lifetime of the device, or you can modify it as needed.
Each module indicated by arrows is a different kind of card with different interfaces. The one at the bottom has 48 SFP ports, the next two are 48 RJ-45 Ethernet ports each, the next ones up may be the two RSP cards which I’ll talk about shortly. At the very top you have eight power units which function as redundant power supplies. You can mix and match the various cards as needed. This particular model has 7 slots plus 2 for power, but others may have more or fewer.
An empty chassis looks like this (with covers on the slots):
It is composed of the backplane along with a fan tray (which is what you see on the left). There is no intelligence. In order to provide intelligence, you must purchase an RSP or what is also known as a supervisor engine. There are a whole series of supervisor engines that you can purchase with varying capabilities, memory, and processing power. (There are 15 supervisor card models to choose from for the 9400 for example.) In some cases you may want two supervisor engine cards to allow for redundancy, or just to ensure you have enough CPU and memory resources to deal with the expected traffic and network services.
Because the architecture of a chassis switch is a little bit different than a non-modular switch, you have two types of CEF. We have centralized CEF, which uses the intelligence in the RSP cards to perform CEF. Remember, the RSP is not just a CPU, but it is all of the circuitry (CPU, memory, ASICs, CAM/TCAM, and other high-speed hardware) required by a switch. On the other hand Distributed CEF bypasses the RSP by using hardware that is built-in on board each line card. In other words, some intelligence and high speed hardware is included in the line cards themselves. This further speeds up the CEF process since it bypasses the route processor.
You must think about a chassis switch in the same way as a non-modular switch, except that the various components of the switch are separated onto different components (line cards and supervisor engines) for the purpose of modularity. The special case of distributed and centralized CEF is simply because of this distinction between these components. More info can be found at this datasheet of the 9400 chassis switch. If you’re interested, I believe a quick read of that document will further clarify any questions you may have.