use CAT3750 as a core switch

[ronshuster]

Is it recommended to use CAT3750E (2 of them stacked) as a core switches? We have under 200 users in the building but all production servers are elsewhere. The only traffic going from this building is Internet and our private network. There isn't much traffic going through this set of cores.

[Giuseppe Larosa]

Hello Roni,

if the two C3750 stack fits your current needs and mid term requirements you can use them.

Before performance issues, you can face some features limitations for example you need an external device to perform NAT.

Hope to help

Giuseppe

[ronshuster]

Giuseppe,

We have an ASA facing the Internet so there is no issues with PAT'ing. I am mainly worried about the capacity.

[Jon Marshall]

Roni

Considering that your servers are remote across the WAN then the 3750E switches should be fine as your core switches as the limiting factor will most likely be your WAN connection.

Unless you have an awful lot of local traffic which it doesn't sound like you do then you are okay.

Jon

[ronshuster]

Thanks Jon, that makes sense.

We have about 5 other 3750's serving as access switches and a C2821 facing the private network + an ASA facing the Internet and the stacked 3750 will be used as a core. Thanks for the advise.

[Jon Marshall · Accepted solution]

No problem, glad to have helped.

[Joseph W. Doherty]

For only 200 users, and if most traffic it bottlenecked elsewhere, a pair of 3750Gs might serve too.

A single 3750E is practically a wire speed switch. In a stack configuration, you're "limited" by the stack ring bandwidth (dual 16 Gbps). To avoid this bottleneck, if it's a concern, try to arrange all normal unicast data flows, within the core, to not transit between the dual stack 3750E members (except in a failure).

PS:

One issue, I recall with 3750s, was a limitation on number of HSRP groups (32?). Don't know if small number of HSRP groups also true on 3750Es.

[ronshuster]

Joseph,

What do you mean by "To avoid this bottleneck, if it's a concern, try to arrange all normal unicast data flows, within the core"

Also, there is no HSRP setup as the two switches are stacked and are logically one. The vlan interface is defined on the stack and not on a 1 particular physical switch

[Joseph W. Doherty]

"What do you mean by "To avoid this bottleneck, if it's a concern, try to arrange all normal unicast data flows, within the core" "

I mean the ring bandwidth can become a bottleneck for communication between the two switches. Unlike StackWise, StackWise+ doesn't flood the ring with all unicast traffic. It only forwards unitcast traffic on the ring that needs to be sent to another stitch on the stack.

"Also, there is no HSRP setup as the two switches are stacked and are logically one. The vlan interface is defined on the stack and not on a 1 particular physical switch"

I had assumed that, but for instance, using HSRP is one way to avoid the gateway MAC changing if the stack master fails. In other words, there are still reasons for using HSRP.