Why do we need a 'core-layer' ?

adel.g.qod

This questions is really bugging me.

So, I have my vlans, they're connected to routers which do the inter-vlan-routing and these routers provide WAN connections, ACL and other things.

So where does the Core-layer come in, here? If the routers are actually connecting my vlans, and they're connecting branches together, and they're connecting with each other, why would I need a core-layer?

I do realize Core-layer exists for a reason, I just don't quite get that reason yet :P

P.S Sorry I am posting this in the CCNA section because I found the Three-layer-architecture in my CCNA book
P.P.S Excuse my ignorance, I know :P The question might be very primitive

oli356

Access - desktop access etc. things like ACLs are used, routing protocols are used here (typically the ones with higher convergence times), few redundant links, typically 100Mbps per switchport and 1Gbps or 10Gbps to distribution layer. 13:00

Distribution - More ACLs to control traffic movement and restrictions, QoS is used here, routing protocols with a lower convergence time are used here, higher levels of redundancy through redundant links/devices )(HSRP/Stacking/VSS). 13:02

Core - Switching only, no intensive ACLs, QoS or firewalls are here. its purely high throughput, high bandwidth switching. HIGH redundancy, duplicated devices, many redundant links. links to the WAN or other sites

----
Someone sent me that a few days ago. Hope that helps a bit.
On a home lab though you won't be using the 3 layer model, it's for large networks. On the Cisco site you can see core and distribution switches as well as access layer ones.
http://www.cisco.com/en/US/products/hw/switches/index.html the amount of data those things can send per second is incredible.

networker050184

The three layer architecture is just a model and all networks do not have to follow it, especially small networks like you are describing. If you have a very large network with hundreds or thousands of devices its a good idea to have a high speed core layer set up that can handle all of the traffic from your different distribution sections. If you have a couple switches and a few routers then it makes more sense to go with a collapsed core/distribution design. Hope this clears it up a bit and not makes it more confusing for you.

adel.g.qod

@Oli:
Thanks, I've read this before though.

@Networker:

Okay, fair enough.
Let's say I have a large network, with thousands of hosts:
Still, my lans/vlans can be connected to routers which in turn connect with each other via dedicated links, why would I need to have a set of dedicated switches to link these routers for me? I am sure there's a reason, just bear with me :P What is that reason?
Currently, it only looks like I am making a bottleneck.

Illumanati

This thread was very informative for me and I wanted to thank the author for posting it. this was just "supposedly" covered in my Cisco IV class but no where with the clarity as these responses!

johnifanx98

oli356 wrote: »

Access - desktop access etc. things like ACLs are used, routing protocols are used here (typically the ones with higher convergence times), few redundant links, typically 100Mbps per switchport and 1Gbps or 10Gbps to distribution layer. 13:00

Distribution - More ACLs to control traffic movement and restrictions, QoS is used here, routing protocols with a lower convergence time are used here, higher levels of redundancy through redundant links/devices )(HSRP/Stacking/VSS). 13:02

Core - Switching only, no intensive ACLs, QoS or firewalls are here. its purely high throughput, high bandwidth switching. HIGH redundancy, duplicated devices, many redundant links. links to the WAN or other sites

----
Someone sent me that a few days ago. Hope that helps a bit.
On a home lab though you won't be using the 3 layer model, it's for large networks. On the Cisco site you can see core and distribution switches as well as access layer ones.
Switches - Main Page - Cisco Systems the amount of data those things can send per second is incredible.

It looks like core-layer refers to WAN technology/devices, specifically about layer 2.

adel.g.qod

johnifanx98 wrote: »

It looks like core-layer refers to WAN technology/devices, specifically about layer 2.

It doesn't seem so; according to my CCNA book(Lammle), the Distribution layer is the one responsible for WAN connectivity.
The Core layer's main responsibility is to 'Switch traffic as fast as possible'.

spiderjericho

If you go buy the book (CCNA, CCDA, CCDP, etc), it depends on your network and obviously none of these books mention Trill, SPB, virtual routers or discuss the latest Cisco Nexus or Catalysts.

But say you have a campus with six three story buildings. On each floor of each building you have a Nexus 6500 switch servicing 200+ users. So for each switch on each floor, they're functioning as an access switch.

Now, you also have a 4500 switch on the first floor of each building that aggregates all the trunks from the 6500s and connects the building to the rest of the campus. These would be your distribution layer and possibly strictly uses fiber.

Then to connect all six buildings together and to your routers, you have a couple switches, maybe 6500s or Nexus 7000 acting as your core layer.

It just all depends on the size of your network and use case.

adel.g.qod

Great example, thanks!

spiderjericho wrote: »

If you go buy the book (CCNA, CCDA, CCDP, etc), it depends on your network and obviously none of these books mention Trill, SPB, virtual routers or discuss the latest Cisco Nexus or Catalysts.

But say you have a campus with six three story buildings. On each floor of each building you have a Nexus 6500 switch servicing 200+ users. So for each switch on each floor, they're functioning as an access switch.

Now, you also have a 4500 switch on the first floor of each building that aggregates all the trunks from the 6500s and connects the building to the rest of the campus. These would be your distribution layer and possibly strictly uses fiber.
Cool, so the 4500 switches got connected together as you say, using fiber.
What else is needed? There, you got them connected, why would you need the 6500s or Nexus 7000 acting as a core layer?
Then to connect all six buildings together and to your routers, you have a couple switches, maybe 6500s or Nexus 7000 acting as your core layer.

It just all depends on the size of your network and use case.

Again, I am really sorry, just bear with; I seem to be missing the point

spiderjericho

Don't get lost in the Cisco technology portfolio. You can use any type of switch for any layer.

The reason why I used the Nexus 7000, 6500 or 4500 is port density.

A 3560X, 2960 or 3750X can only have a maximum of 48 ports. If you use my example of having 200 users, you'll need a switch with a lot of ports like the 6500.

From each building's distro switch you're probably going to use fiber to connect to a core switch. For redundancy, you might have two of these.

Again, ignore the names of the switches. The best way to "see" this is probably making a quick network in packet tracer with a bunch of access switches connected to PCs and think to yourself, how do I connect all of the users behind the switches together? You can't physically interconnect the Access switches together, you need a switch to aggregate the trunks.

NetworkVeteran

OP wrote:

Cool, so the 4500 switches got connected together as you say, using fiber.
What else is needed? There, you got them connected, why would you need the 6500s or Nexus 7000 acting as a core layer?

You've received many responses about the role of the core layer, and what real networks look like, but none directly tackling your question. Before I do that, I want to point out that the CCNA/CCNP/CCIE are about implementation and troubleshooting. If you want to learn about design, I encourage you to pursue the CCDA/CCDP/CCDE.

Now, suppose your campus has only four buildings (eight distribution-layer switches). A full mesh between distribution-layer switches requires 64 high-speed ports and running cables from every building to every other building. A core model would instead require 40 high-speed ports and running cables to only one central place. Already, you can see the core model is cheaper, easier to manage, and easier to scale.

No-Core Ports: n^2
Core Ports: 5n

If your campus is medium-sized the problem becomes exascerbated. Imagine you have eight buildings (sixteen distribution-layer switches). A full mesh requires 256 high-speed ports vs. a core model that requires 90 high-speed ports!

There are additional concerns here. For example, you will be using 16 high-speed ports on each distribution-layer switches in the first model vs. 3 high-speed ports on each distribution-layer switch in the second model. The 4500 isn't really intended for such a role and "only" supports 48GB of bandwidth per slot. This could be a problem in some scenarios, requiring you to upgrade all sixteen of your distribution-layer switches!

We're only scratching the surface here. Food for thought.

EDIT: I meant 48Gb, not 48GB. Capitalization matters, sometimes!

adel.g.qod

@NetworkVeteran:
Ah that's it!
I kinda feel dumb for not seeing this :P It all makes sense now; building a mesh as the network grows will be increasingly harder/more expensive/less scalable.

As for the CCNA: Thanks for the advice, my intention for going CCNA(and then CCNP R&S) is merely because a lot of people told me that I'll start as an Admin and then move on to a Designer/Engineer role, so it made sense I start with CCNA first.


Edit:

NetworkVeteran wrote: »

The 4500 isn't really intended for such a role and "only" supports 48GB of bandwidth per slot. This could be a problem in some scenarios, requiring you to upgrade all sixteen of your distribution-layer switches!

A question, what do you mean by "Supports 48GB of bandwidth per slot" ?

NetworkVeteran

Glad I could help.

I'll start as an Admin and then move on to a Designer/Engineer role, so it made sense I start with CCNA first.

That would be a typical progression.

adel.g.qod wrote: »

A question, what do you mean by "Supports 48GB of bandwidth per slot" ?

Internal to each switch is a network of sorts. When you buy switches, it's important to note both the chassis a.k.a the switching fabric bandwidth, the bandwidth per slot, and the bandwidth per card. Buying a 12-port 10GE card doesn't guarantee 120Gbps of bandwidth! Devices marketed for the core often get very close to achieving linerate fully-loaded.

Cisco Website wrote:

The Cisco Catalyst 4500 Series offers two classes of line cards: E-Series and classic. E-Series line cards provide 24 or 48 Gb of switching capacity per slot. Classic line cards provide 6 Gb of switching capacity per slot.