Ethernet switch redundancy

A

ashman

Gawd
Joined
Mar 28, 2011
Messages
811
I'd like to introduce redundant ethernet switches at a few of my clients. I did this about 15 yrs ago with 3Com switches but cannot remember how I did it.

For example, right now, all the drops come into one 48 port managed switch, how do you implement redundancy in such as way that, should that switch go down, everything keeps running?

Not sure what is involved in terms of budget, not there yet, just really need a solid understanding of how to best implement a redundant solution.

Thanks.
 
You should be able to double up on switches in most cases if you really need to. STP (Spanning Tree Protocol) should work automatically to prevent loops, broadcast storms, etc
 
So explain how a drop that connects to port 1 on switch A, can be made redundant by doubling up on switches? Does it work something like, port 1 on switch A is mapped to port 1 on switch B, if so, how does the physical cabling work?
 
Hi ashman,

I think you might be confusing 2 different concepts.

Spanning Tree Protocol (STP) is a protocol that was invented to prevent layer 2 loops in switches (broadcast storms). This permits net admins/engineers to have redundant switches branched together. Here is a basic diagram of STP:

ciKNjak.png


What you might be talking about is a mesh topology as far as physical connectivity. Meshing is expensive because it requires at least double the amount of cabling, if you are talking about to end points.

Meshing and spanning tree protocol go hand in hand when you are talking about your switching infrastructure.

If on the above diagram, say you had a host connected on Switch A (Host A), and another host connected on Switch F (Host F). These are all on the same broadcast domain (VLAN). In the even that switch C were to go down, Host A would still be able to reach Host F through Switch D.

STP can get pretty involved and complicated when you expand beyond something as simple as this, but if you are just talking about having multiple access/distribution switches connecting to a redundant core switch STP will determine the quickest way by establishing cost, etc.
 
Ok, so your saying the only way to have redundancy in the manner that I described, is to have two ethernet runs to every workstation? Seems kind of crazy but if thats the only way to do it.
 
That is if you want redundant links to an end workstation. That would only be necessary to mission critical computers.

If you wanted two switches connected, you can theoretically uplink them with 2 ethernet cables. Most L2 switches that are worth a salt have STP enabled by default. You will know immediately if this isn't working, as the lights on the switches will be going nuts and traffic will halt.

But yes, if you want redundancy on your LAN from host->host you would need to mesh everything...
 
Really what I want is, when a switch dies, not have to physically unplug every cable going into the current 48 port switch and plug them into a new one, but I'm guessing there isn't a more elegant way to do it. Can't some ethernet switches connect at the back so that they communicate for redundancy?
 
You are talking about stackable switches. But this doesn't solve the physical connectivity of your hosts problem.
 
So in other words, even with stacked switches, if the primary fails then, everyone is still down? What if I had three switches with all the hosts divided into two of the switches with a third for failover, would that work?
 
How can that possibly work if the switch on which the hosts are connected have failed?
 
If you want end points up when a switch fails, you need to connect all endpoints to more than one switch.
 
OK explain the architecture, because that is what I'm after, unless the diagram in the link for A/B failback switch is the only way to do it.
 
Easiest way might be to just split up connections across two switches, make sure each department has computers connected to both. Should one fail at least there will be some computers that still work in every department.

I wonder what would happen if you had two switches configured the same way, but one is turned off, and if you had some kind of ethernet Y cable going to both from the patch panel for each jack. I think that could technically work, but if they are accidentally both turned on at once, I'd hate to see what happens. :D

Switch failures are not really something that common though, unless it's Netgears. :p You are probably more likely to have a UPS failure than a switch failure, so if you go to this length you should probably have two UPSes too.
 
This is not a common point of redundancy in an enterprise. Switch failures are uncommon and the impact small enough that it really does not matter.

Servers typically have multiple connections to multiple switches due to their mission critical nature.
 
Grentz said:
This is not a common point of redundancy in an enterprise. Switch failures are uncommon and the impact small enough that it really does not matter.

Servers typically have multiple connections to multiple switches due to their mission critical nature.
Click to expand...

Cmustang87 said:
Exactly what Grentz said.
Click to expand...

This. If the worst part of my day is installing a 48 port switch I need a new gig!
 
I ran across this, registered to post on this forum, just to give my 2 cents...

There are SEVERAL ways to do this, they've been mentioned by other posters already.
1. "Fallback switch" is one way, "on the rack" as was described. That means, I assume, a single run of Cat5/6 cable from your end user's PC to the closet where the switch is. In the closet, some kinda device (hopefully passive) splits that run to 2 fallback switches. If one fails the other one "takes over".
2. Dual NIC's in the PC, dual runs to a closet, with two switches. The NIC's on the PC use "teaming software" to create an etherchannel to 2 different switches. This is the "industrial strength" way to go, it's expensive, it's more appropriate for servers in a datacenter.
3. Gateway Redundance Protocol is a default gateway redundancy, which might be what you're really after--although I thought you explicitly said you were concerned about a SWITCH failure. Gateway Redundancy would let 2 different routers act as a default gateway for all your PC's. If a router failed you'd be covered. If the actual LAN switch failed, this wouldn't help you. But, hey, if you're that concerned about redundancy, then you need this, too.
4. Put a wired NIC, AND a wireless NIC in each PC and hang an access point. As cheap as wireless access points are these days, this seems like a great alternative to me.

Option 1 is good because you only have one run of cat 5/6 cable to the end user's PC from the closet. BUT, I don't know what those switches cost, what features they do/don't have that you may also need (voice VLAN, QoS, plain old VLAN's, PoE for phones, security featuers like DAI, CoA for Radius/NAC, etc).

Option 2 is hands down the way to do this--if you have servers. Teaming software on an end user PC...I don't know, maybe not. This option will be expensive because the switches will need to support what's called "Multi-Chassis LAG", or "Multi-chassis channel", or "Virtual Port Channel". That means high end switches in the closet. Don't often see that. Switches like that typically go in the DataCenter to connect servers that can't ever go down. This will also require a second "wired NIC" in the PC, AND a second run of cable to that NIC. Most everywhere I go, people are trying to stick with ONE run of cable to the PC/IP phone. Nah...this option is more for the datacenter and servers.

Option 3 is not exactly what you want, but if you're gonna take the time and effort to provide switch redundancy for end user PC's, then you need to also take the time to provide gateway redundancy for them, too. Look up GLBP, VRRP, HSRP. 2 different routers will answer for a virtual address. You put that virtural address in the PCs' "default gateway" setting, and they will survive a router failure, too.

Option 4 is, most likely, the best way--cheapest and easiest is what I mean when I say "best". But that may not be best depending on your priorities. But, my laptop does this at my house right now. My wireless card is connected "automatically" when I walk in the house. If I plug in a Cat 5 cable, that is what is used to get on the network (I think). If one or the other disconnects from the network, I don't drop a ping. I didn't even TRY--that's just how it works.
 
Last edited:
OP, if you are running into a lot of issues where end user access switches (or "closet" switches as we like to call them) are failing, then you are better off at looking at why they are failing in the first place. Cheaper/low quality brand? Environmental conditions? Power? - these are all factors that need to be looked at.

From an architecture point of view, everyone has already mentioned the only way to do it and that is dual runs to every desk going into redundant, enterprise class switches. Even in this use case, you really wouldn't be using this type of switch properly as they are designed to run LACP between multiple chassis - common desktops/laptops don't support LACP except for maybe some Linux boxes. A cheaper option would be to use stackable switches but you still incur the high cost of multiple runs to each desk.

Cheapest way would be to have a solid WiFi infrastructure running off of separate equipment or getting the proper "closet" switch gear in the first place would even be better. If you go the WiFi route, you have to do it right in that it's sized and designed properly for the area you are looking to cover as well as being able to handle the throughput if every user were to jump onto it. There is expensive software such as Ekahau that can help you design this but it's useless if not used properly by an engineer who knows what he/she is doing.
 
You must have shitty switches or power if you are that concerned about failures. I have Cisco 3550s from 2004 happily purring along with no issues.

Buy two switches, keep one as a cold spare.
 
You must log in or register to reply here.
Back
Top