Building redundancy

[bflseanny]

When buidling redundancy into the network, it is often recommended that an access layer swith have seperate connections to two (or more) distribution layer routers.

In this design, it seems that the assumption is that the router is more likely to go bad than the switch port connecting to it, correct?

What is a good method of providing redundancy at the access layer besides dual-homing every device on the network to two differenct access layer switches?

[Jon Marshall]

Sean

"In this design, it seems that the assumption is that the router is more likely to go bad than the switch port connecting to it, correct?"

Not really. The assumption is that each switchport and each router is equivalent ie. either one could go and there is another to take it's place.

However the switch as a whole is still the single point of failure - see below.

"What is a good method of providing redundancy at the access layer besides dual-homing every device on the network to two differenct access layer switches?"

You generally don't do this anyway. Most devices are singly honed ie. end hosts so if the switch goes so does the device. I have not come across a network where end hosts are dual honed. So having 2 separate switches with half the hosts on one and half on the other, and one of the switches goes, you have then lost half the clients.

You can mitigate this with either 3750 stacks or modular switches such as the 4500/6500 where you can deploy redundant supervisors.

Servers are another matter and these are generally dual honed although they are usually in their own server block rather than the general access-layer.

Jon

[bflseanny]

So, the switch will always be a single point of failure for the end hosts, correct?

In other words, even if I have stacked switches, if that switch that my hosts are connected to goes bad, I would still have to reconnect their patches to another switch in the stack.

Just looking for wisdom...Thanks in advance.

[vmiller]

Thats pretty much it. Its a cost/benefit/complexity/probability question. Some times its better to have a predictable failure sceneario than over complicate redundancy and not know whats happening.

[Jon Marshall · Accepted solution]

Sean

"So, the switch will always be a single point of failure for the end hosts, correct?"

If the switch dies then yes ie. if the switchport to which the host is connected dies then the host loses connectivity. However it depends on which bit of the switch dies eg.

a 4500 switch with dual supervisors and dual power supplies. End host is connected to a single copper port on one of the line cards.

If the line card or the port on that line card dies then the host loses connectivity.

If another line card within the chassis dies this does not affect the host.

If a supervisor dies then again this does not affect the host as there is a redundant supervisor.

If a power supply fails, again this shouldn't affect the host unless you are running power supplies in combined mode in which case it may or may not affect the host.

So you can see that there is quite a lot of redundancy built into a 4500 switch. And of course there is a cost to go with that.

Stacked switches are slightly different. The 3750 has StackWise technology will allows for one of the switches in the stack to be a master and the others slaves.

Obviously if the switch you are connected to dies then the host loses connectivity. But if the master dies and you aren't connected to the master switch then one of the slaves can take over in which case your host will not lose it's connectivty.

Basically as you say for end hosts that are singly honed the switch is always going to be a single point of failure. However with modular switches and 3750 stacked switches you can build in some redundancy to the chassis's themselves.

Jon