Does Stackwise provide redundancy?

[cbalowski]

Hi,

Let's say I have a stack of two 3750s, 24 ports each (48 ports total).  The backplane is stacked correctly.

I also have 20 Servers and switches which need to be availble to network users 24x7x365.

How do I plug these 20 servers and switches into the stack to provide High Availability in case the one member of the stack fails?

Can some one describe how StackWise provides 'redundancy'?

Thanks,

cbalowski

[Jon Marshall]

Simply connect them to both switches in the stack.

Your servers would have 2 NICS which could be run in fault tolerant mode or you could etherchannel them to the switches. Your switches would be connected to both stack members as well with an ethechannel link.

The good thing about a 3750 stack is that you can spread the same etherchannel across both 3750 stack members.

Then if one of the 3750s fails there is still a switch available with the servers and other switches connected.

Jon

[cbalowski]

"Then if one of the 3750s fails there is still a switch available with the servers and other switches connected."

Thanks Jon.

So, using etherchannel, if one member of the stack fails, then the server/switches will still be connected at 1/2 bandwidth, correct?

What about plugging all server/switches into the one member of the stack and, if that member fails, then physically move all server/switches into the other member of the stack?  What are the ramifications (config, speed, etc...)?  Is the running config 'shared' that way?

Also, these 3750Gs are over 7 years old and running IOS 12.1(11)AX.  Does that image support etherchannel across stack members?  If not, how should I go about upgrading the image on these switches...they do not have SmartNet.

[Jon Marshall]

Yes you would lose half the bandwidth.

Well yes you could manually reconnect but that is kind of missing the point about having stacked switches. You said you needed 24x7x365 which doesn't really happen if you have to manually move connections across

As far as i know the 3750s supported MEC (Multichassis Etherchannel) from the start. If you need to upgrade but don't have a contract for upgrades then there is probably a cost involved. You would need to speak to your local reseller. 

Jon

[cbalowski]

"Well yes you could manually reconnect but that is kind of missing the point about having stacked switches"

I see, but its only kinda missing the point if I'm able to be alerted and respond to the failure within a reasonable timeframe, say 3-5 minutes. Therefore the uptime would be 24x7x365 - 5 minutes : )

Anyway, I'm interested in how the running config adapts to the change in the above scenerio.

Lets say that I have several VLANs and some routes setup on the primary/master 3750 and that I manually failover (move all of the CAT5 cables off of the dead master to the same ports on the standby member), would the running config of the stack handle/allow that automatically? Do I have to reprovision anything in the config?

[Jon Marshall]

Sorry, wasn't intended to be a dig, it's just that if the switch goes overnight and for example you had users who remotely logged in or early morning users and you are stuck in traffic etc.etc. then i could forsee a lot longer than 5 mins.

The config should be sent to the other switch so you should just need to reconnect but again i would strongly recommend using both switches for connections if you need anything like 24x7x365.

Edit - actually you may need to do additional config because the switchports would need assigning into the correct vlans and this wouldn't exist in the current config although you could configure it for that eventuality.

Jon

[cbalowski]

Thanks Jon...no offense taken.

This is where I am confused do to lack of knowledge.

What is wrong with my thinking, below:

1) A stack of two 3750s (24 ports each) appears to the StackWise IOS as 48 interfaces.

2) StackWise IOS sees that member one (master) has interfaces 1 - 24, and member two (standby) has interafaces 25-48.

3) All 20 servers/switches are plugged into interfaces 1 - 20.

4) Member one (master) goes toes up. Interfaces 1 - 24 are dead.

5) StackWise therfore 'remaps' interfaces 1 - 24 to the standby.

6) Now, what happens to interfaces 25 - 48 of the standby???

How is VLAN, routes, etherchannel, etc... that is configured for interfaces 1 -24 remapped to interfaces 25 - 48?

[Jon Marshall]

No the interfaces are not remapped, they keep the same numbering so you would be okay as far as that goes.

Vlans you could just add the unused ports into those vlans ready for reconnecting and routes would hopefully be by next-hop rather than interfaces.

Jon

[cbalowski]

"As far as i know the 3750s supported MEC (Multichassis Etherchannel) from the start."

Looks like I have the minimum version 12.1(11)AX which supports cross-stack etherchannel.

I just referenced the manual that came with the switch

Still confused about the replication of the config file and interface management across stack members.

OK

A stack of two 24port 3750s has 48 unique interfaces, gigabitethernet 1/0/1, gigabitethernet 1/0/2.............gigabitethernet 2/0/1......etc.....right?  If gigabitethernet 1/0/1 fails, and then I move the CAT5 cable from gigabitethernet 1/0/1 to gigabitethernet 2/0/1/, how does gigabitethernet 2/0/1 assume the identity (name, config, etc...) of gigabitethernet 1/0/1?

What am I missing?

[Jon Marshall]

OK

A stack of two 24port 3750s has 48 unique interfaces, gigabitethernet 1/0/1, gigabitethernet 1/0/2.............gigabitethernet 2/0/1......etc.....right?  If gigabitethernet 1/0/1 fails, and then I move the CAT5 cable from gigabitethernet 1/0/1 to gigabitethernet 2/0/1/, how does gigabitethernet 2/0/1 assume the identity (name, config, etc...) of gigabitethernet 1/0/1?

What am I missing?

No it doesn't assume the identity of the gi1/0/1. This is what i meant when i said when both the switches were up and running you would need to config the gi2/0/x interfaces the same way as the gi1/0/x interfaces ie. assign them to the same vlans, configure other interface settings etc

Basically you need the gi2/0/x to mirror the gi1/0/x interfaces in the config. Then if the gi1/0/x switch fails with the actual connections in it you can just repatch to the equivalent gi2/0/x interface.

The rest of the config ie. routing/vlan database/L3 SVIs would automatically be on the new active switch so you don't need to do anything there.

Note you would only have to do this if you went with the setup where you only connect servers/switches to the one switch. If you use both switches for connectivity then you should not need to do this.

Jon

[Leo Laohoo]

Sorry Jon.  I don't want to step over your toes here ...

So, using etherchannel, if one member of the stack fails, then the server/switches will still be connected at 1/2 bandwidth, correct?

You have two Cisco 3750 stacked together. Each stack member has at least ONE uplink.  If this is your setup then, as Jon has mentioned, you will loose half your switch stack but your server, with two NICs per server and each server has one connection to each of stack member, then you only loose half your NIC.

What about plugging all server/switches into the one member of the stack and, if that member fails, then physically move all server/switches into the other member of the stack?

This ain't "redundancy".  Redundancy that requires human intervention isn't redundancy.

Also, these 3750Gs are over 7 years old and running IOS 12.1(11)AX.  Does that image support etherchannel across stack members?

Wow!  That's pretty old.  Don't know if 12.1 supports Etherchannel.  I presume they do but the algorythm is pretty ancient.

Lets say that I have several VLANs and some routes setup on the primary/master 3750 and that I manually failover (move all of the CAT5 cables off of the dead master to the same ports on the standby member), would the running config of the stack handle/allow that automatically? Do I have to reprovision anything in the config?

As long as each stack member runs the same IOS version and the same feature set, the configuration is "shared" exactly the same.  This is due to the redundancy feature of the 3750.

2) StackWise IOS sees that member one (master) has interfaces 1 - 24, and member two (standby) has interafaces 25-48.

That's not correct.  The stack is composed of one "module" as Gig 1/0/1 to 28 (1-24 copper and 25-28 as SFP) and another "module" as Gig 2/0/1 to 28.

3) All 20 servers/switches are plugged into interfaces 1 - 20.
4) Member one (master) goes toes up. Interfaces 1 - 24 are dead.
5) StackWise therfore 'remaps' interfaces 1 - 24 to the standby.
6) Now, what happens to interfaces 25 - 48 of the standby???

If all of your server's NIC are plugged into Gig 1/0/1 - 24 and if switch #1 goes down, then I hope you have a teleporter machine because none of your servers are going to pass traffic until human intervention is made.  This scenario doesn't fit the design of the 3750.