Catalyst 2960 vs 3750

[mikecel79]

Long time reader here but first time poster. I am looking into upgrading our network switches soon and I'm trying to decide on a design now. Let me give some background information on the network.

Currently we have a Catalyst 5513 with a Supervisor II, ten 24 port 10/100 cards and a 12 port fiber card installed. There are also two 48 port Alcatel switches that are uplinked to the 5500 as well as six Catalyst 2900XLs (on the factory floor) uplinked to it via fiber. There are about 40 servers and 350clients in total in the building. All servers and most of the clients in the office are connected directly to the 5513. We will be consolidating buildings soon and plan on adding another 200 devices to the network (office and factory floor area).

I was thinking we should get two Catalyst 3750G-48TS and a 3750G-12S in a stack for the core switch. Then use Catalyst 2960-48TC-L switches uplinked to the 3570 stack for the access layer. Does this make more sense than directly connecting all the clients and servers directly to the 3570 stack? What would be the advantages/disadvantages of each setup besides cost?

Thanks for any help you can provide!

[dcrown]

I tend to go with a more modular approach and hang nothing off my core except otehr switches. Put your servers on their own switches as well. There are a few reason i do it this way.

The first is I run two cores for redunacy. by giving servers their own access layer switch, i can isolate them from a core failure.

Second, ports on a core switch tend to be more expensive. I know I'm going to be adding servers in the next few years. Its cheaper for me to buy more 3560's then more blades for a 6500.

Third, modularity. I can make more changes to either the core, or edge switches and not effect the other. Say you find out down the road, that the 3750 stack isnt quite moving the number of packets you need. If you plug end users and servers into your core now, just picture how much of a bear it will be to replace the core. Always try to go the route that will give you maximum flexibility and growth.

[lgijssel]

A network of this size should have some redundancy in it. This can be achieved by connecting server adapters in teamed mode to different switches and also by providing a dual uplink for the access layer. All of this should be done utilizing gigabit ports. When the core is sufficiently redundant it is more economic to directly connect your servers to it instead of using an extra access layer.

However, this means that you might need a bigger stack as the core altough in my opnion, you are coming pretty close to the point where a decent 4500 series switch would be a better choice. You can still use the 2960 series in the access layer.

Regards,

Leo

[mikecel79]

Thank you both for your input. I have revised our layout to provide redundancy in the core as you suggested and also adding an access layer for the servers. We will probably have an increase in the amount of servers over the next few years so adding an access layer allows us to be more flexible.

Thanks again!

[dcrown]

When planning out your core redundancy, You could consider not just hard ware and cabling, but what protocols you plan to run. Consider revisting your spanning tree, and routing protocols. For an investment in your time, you can have some phenoinal reduncancy and reconverange time.