Design for Redundant Links?

January 13, 2009 at 12:45:04
Specs: xp, p4 2m
How can one configure a LAN to have redundant links to a single NIC or MAC address? This would be called fault tolerant so that if one link failed the other link would still provide data. This would be 2 100BaseT links from a router/switch to router/switch and to a single NIC/IP address.

Would duplicate data be sent on each link and the router somehow select one or would the router reconverge to the redundant link by sensing a link failure and reconveging to the backup?

Any thoughts or literature you might refer me to so that I could understand how to do this or if it can be done?

Thanks for any tips

See More: Design for Redundant Links?

Report •

January 13, 2009 at 13:15:04
It can't really be done but you can get close in a business world [spend lots of money]

For example I have dual nics in the server that are adapter teamed. This means if one nic goes down net access doesn't go down.

This connects to a Nortel 8010 passport switch. One nic connects to one blade and the other nic connects to another. This unit has two Switch fabric modules [the brains] along with three power supplies.

This means I am still operational if I
lose 1 nic
lose 1 blade
lose 1 SFM
lose 2 power supplies.

I only have one fiber connection between this backbone switch and all other downstream switches. You could do two connections to each switch if managed switches and enable STP [spanning tree protocol] which disables the second link until such time as link 1 goes down.

We didn't so this due to fiber wiring of building cost and have had no need for redundancy at this level.

A soho or standalone router has no capablities unless it was a dual wan router and that only pertains to internet access not lan access.

You can get more with a layer 3 switch and you could build router redundancy that way but my Passport cost $78,000 nine years ago. If that gets you an idea of what cost we are talking about here. A single replacement blade can cost between $6,000 and $10,000. I don't even want to know what a SFM would cost.

Usually the best you can do is adapter teaming to a supporting managed switch.

Report •

January 13, 2009 at 13:30:28
I'll point you in the right direction. What you're looking for is either: MLT or SMLT

MLT = Multi Link Trunk
SMLT = Split Multi Link Trunk

However, you said, "How can one configure a LAN to have redundant links to a single NIC or MAC address? "

The problem with your scenario is, you cannot provide redundancy to a single NIC because if it fails, or the device it's a part of fails, then you lose your connection....even if you have an MLT or SMLT configured.

Take where I work for example. We have dual 8600 Nortel Passport core switches as the heart of our switched network. They are connected by dual IST's (inter switch trunk) which allow them to communicate with each other and provide redundancy. Should the primary fail, it switches over to the other one so fast, you don't even notice a hiccup on the network.

Connected to the core switches are our stacks of switches. I'm only going to discuss the "client stack" that consists of 5 Nortel Baystack 5510 switches connected via cascade cables. This allows them to behave as one big single switch.

The client stack is connected to our core switches (Core A and Core B) via an SMLT. Because of the way this equipment works together, the two links aggregate bandwidth so that I have 2 GB's of bandwidth between the client stack and the core switches.

One half of the SMLT connects to Core A, the other to Core B. This provides redundancy. Should Core A (the primary) fail, all data travels to the client stack from Core B. Granted, I lose 1 GB of bandwidth, but the remaining 1 GB is still an awful lot.

Anyhow....on the client stack I have port 48 on unit 2 as one uplink and port 48 on unit 4 is the other. Should unit 2 fail....all remaining units are able to continue communicating via the uplink in 4/48. Again, bandwidth drops to 1 GB but, the fact that all other switches, but the broken one, can continue to communicate means we have redundancy in both directions (ie: broken core switch or broken stack member).

So you see, it's nearly impossible to actually build in any redundancy with a single switch connecting to a single switch. At best (depending on the equipment you're using) you can double the bandwidth between switches.

Report •

Related Solutions

Ask Question