to stack Cisco 2960/3750x or to not stack???

[Hokie97VT]

I have competing thoughts of school providing me guidance. My vendor who I purchase all my Cisco gear from recommends stacking of switches where my Cisco Consultant who manages our 6 offices routers and switches does not recommend stacking.

The consultant states that if something goes screwy with the switches the entire logical switch is down bringing the entire site down. However in non-stacked mode, if one goes down I may have some devices down but the other switches are still up. Thus users can still function to some degree.

What does everyone else think??

To stack or not to stack?

 

[VinceWhirlwind]

If he manages them, then it's his problem if he chooses to create extra work for himself by not stacking them.

Personally, I don't see why you would not stack them. What is this "something" that "goes screwy with the switches"?

By stacking them you can get proper redundancy by using etherchannel across multiple members of a stack.

 

[dgrizzard]

Exactly you have redundancy on the back end (stack links) and then utilize etherchannel from each member switch to your distribution/core for another layer of redundancy. Not to mention the management benefits (one change applies to the entire stack).

 

[unclerico]

stack and then get another cisco consultant

I hate all Uppercase... I don't want my groups to seem angry at me all the time! =)
- ColdFlame (vbscript forum)

 

[Hokie97VT]

Just to clear up the implementation. 3 Cisco Switches, these are the primary switches at the site, no core to etherchannel to. No tremendous growth in ports needed at this time. Unless we acquire new office space there would be no substantial growth of ports/devices.

As for redundancy in the stack itself, I am not sure I follow. Just because they are stacked does not mean they are redundant in my mind. What ever devices are plugged into the ports on the physical switch that goes down will still lose connection even if they are a on virtual switch in a stack, correct?

Just playing devils advocate.

 

[Hokie97VT]

I am also curious what every thinks of the RPS2300 to connect 3 2960s for redundancy on the power side vs getting a 3560 with redundant power supplies in the switch itself?

Pros/Cons of each option.

 

[unclerico]

so you'll stack the three switches together and use them as a collapsed core. the beauty of the 3750's is as you grow your infrastructure needs you can grow the stack as well (up to 9) and still manage all of the devices as a single virtual chassis.

as for redundancy, no, you won't see the benefits on an access port for end users or for systems that only connect with a single ethernet cable (the same will be said if you don't stack them). you will see the benefit in systems that have a teamed connection. if you're running VMWare or an iSCSI storage solution you can use x-stack etherchannel and multipathing to get some better availability for your systems.

the RPS2300 is, IMHO, a must for the 3750's and for 2960's (or other access layer switch) that might provide PoE to ip phones, access points, or security cameras. i wouldn't get a 3560 unless you wanted to take advantage of the multilayer capabilities that it provides.

I hate all Uppercase... I don't want my groups to seem angry at me all the time! =)
- ColdFlame (vbscript forum)

 

[vipergg]

Yes if you lose one switch yes you will lose connection for the one switch , if you have dual uplinks then everyone else in the stack will continue to operate , if no dual uplinks then if the switch that went down had the uplink they will be isolated .
As far as the RPS2300 some people get the wrong idea with this , all this device does is backup the internal switch power supply if the internal power supply goes belly up , then the switch will continue to work until you get a replacement internal supply from cisco. This is not a backup power supply , if power goes out then the switch is down , you would have to have a ups behind the rps2300 for power supply for a power failure scenario.

 

[VinceWhirlwind]

Just out of interest - where is the routing occurring on your network?

 

[Hokie97VT]

Why do you ask?

We have routers connecting our sites across the WAN. All local traffic goes through the switches.

So for example one of our sites looks like this:

TLS/ELAN Provider -> Cisco 2800 router -> Cisco 2960G switch -> 2 or 3 Cisco 2960G switches.

 

[Quadratic]

I think your consultant may not be comfortable with stacking, hense the "something screwy" excuse. If he's not familiar with it, he's not comfortible implimenting it and thus would rather stick with what he knows. That's what I suspect, anyway, and if you think about it it's not a terrible reason; the support staff's ability to maintain and troubleshoot a network is an important part of any design.

In this case, though, if he's not familiar then he should read up on it and get familiar. Especially where he's a Cisco consultant specifically, there's no excuse for not knowing or learning how to stack, especially when there's a real-world need for it right in front of him.

CCNP, CCDP

 

[nogoodhandle]

Hi
well since I have some experience with this I would like to offer my view/opinion on the subject.

First off.
Yes your consultant is right, sort of.
He has a valid point of view.
I have had stacks shut down/reload after a software bug.
however that said it is not a something that happens often
(2 times over the last 5 years) and only when testing new software.
and
When upgrading you will (if you are smart) upgrade the whole stack at once. = downtime roughly 5.5 minutes (reload time).

However there are severe drawbacks with connecting them in a daisychain. fx the bandwith available for a packet traversing the stack from one switch to another is alot greater than traversing over a daisychain.

There are also some pros to connect in a daisy chain, such as security reasons, if you get access to one switch then you automatically have not gotten any of the others,

There are always pros and cons when choosing a design.

That said I would go for the 3750x in a stacked environment.
why ?
* power stack feature. (max 4 units)
* New generation of a proven concept
* Stable and works well (if correctly used)
* easier to maintain (less configuration in one switch than 3)

A small clever thing to do for power purposes since you are not using poe. Buy a to big powersupply for the first 2 3750X and then use stackpower to connect all three of them.
and the last one does not need to have a powersupply att all if you are in to saving money.
if you are not and want to have added redundancy then just buy a regular one for it. connect them to different power outages. this means that if somethings goes wrong with the power in one outage then the whole stack can run out of the other. or if you need to move the power cables, then you can do so without interupting services.
works amazing.

stacked is the way to go with all that extra bandwith to traverse packets between units its a given.

If you want to go for extra uptime and availability then you connect servers/switches via redundant links to different hardware switches in the stack.

good luck

 

[teleoptions]

As a Cisco vendor and putting in many switches for end-users, we highly suggest following Cisco's recommendations. We have stacked all of our switches and have never had any problem with that particular set up.

 

[Hokie97VT]

What would be everyones recommendation be if I were to say that the max data moving between the 2 switches is no more than 50 Mbps receive/transmit.

 

[dgrizzard]

I would say move to a Force10 S25/50N solution since it has a 96Gbps stacking capacity compared to Cisco's 64.

I would also say you still get the management benefits and the added redundancy.

 

[VinceWhirlwind]

Regardless of the amount of traffic, stacking gives various benefits and no downsides.

I have installed over a hundred stacks of 3750s and never found any problem that made not stacking them seem like the better option.

I *have* come across badly stacked 3750s causing problems, so follow the installation guide and don't skimp on the last stacking cable to close the loop.