Solved: Probability of failures on NICs, redundancy acess layer

news2010a · ‎05-15-2007

In your experience do you see people implementing teamed NIC's and taking advantage of failover features very often on the Cisco 3750 stackwise switches for example?

I hear stories from multiple network operation folks that NIC failures become more and more rare and increasingly people are not bothering to setup dual NIC's in order to take advantage of server redundancy. In addition, the cost of using an additional port in the switch is also significant.

Right now I need to purchase a Cat 3750 and since currently most of servers in thte production environment do not have dual NIC's available, it seems I cannot make my case to buy the 3750 stack. I think I will buy just a single 3750.

Is that a trend?

Jon Marshall · ‎05-15-2007

Hi

Whilst i can see Paolo's point i would argue that it is not as cut and dried as this.

For example if you have a data centre environment where you have redundant switches, firewalls, load balancers etc and you have a critical database server that only connects to one of a pair of resilient access-layer switches.

Even if the NIC on the server doesn't go down what if the switch does. If you dual hone the database server then if one of the switches goes down you have an alternate path for the server.

A lot depends on the criticality of your servers. Can you say that the cost of a second 3750 switch and a second server NIC outweighs the potential cost of the server being unavailable.

HTH

Jon

View solution in original post

paolo bevilacqua · ‎05-15-2007

Hi,

personally I would not bother on teaming NIC unless the customer insist on it.

I have never seen NIC failures in either host or router/switch in all my career.

What I've seen instead, is drivers/software/configuration problems when hosts have NIC teaming and that is a reason more for which I think my opinion is well founded.

Jon Marshall · ‎05-15-2007

Hi

Whilst i can see Paolo's point i would argue that it is not as cut and dried as this.

For example if you have a data centre environment where you have redundant switches, firewalls, load balancers etc and you have a critical database server that only connects to one of a pair of resilient access-layer switches.

Even if the NIC on the server doesn't go down what if the switch does. If you dual hone the database server then if one of the switches goes down you have an alternate path for the server.

A lot depends on the criticality of your servers. Can you say that the cost of a second 3750 switch and a second server NIC outweighs the potential cost of the server being unavailable.

HTH

Jon

paolo bevilacqua · ‎05-15-2007

Hi Jon,

note I did not say dual-nic hosts, I said teaming-nic hosts. In reality I perfectly agree that key servers benefit in design availability having dual LAN, it's just that I would not bother configuring "teaming", and would be happy with a primary and backup ethernet, on different switches and subnets.

news2010a · ‎05-15-2007

But can you just use both NIC's connected to a Cat 3750 without setting up teaming? I thought that if you don't setup teaming on that one and you attempt to communicate with both switches at the same time you may experience a hell of a loop.

Let me know if I misunderstood your explanation. Basically as far as I know the idea of using two (or more) NIC's on a server implies that one should setup teaming in order to make that one work properly and failover if one of the communication paths (On the NIc itself or the switch) fails.

paolo bevilacqua · ‎05-15-2007

Hi, not necessarily. You can configure each NIC to be into a different vlan / subnet. In this case host has two addresses, I understand that this can not be desired in some cases. With NIC teaming both NICs would be have same IP address, with one blocked by spanning tree. But I have seen application like ms exchange to give problems with teaming, so that is was disabled.

Jon Marshall · ‎05-15-2007

Hi Paolo

Yes, i can see your point but if you address the 2 NIC's out of different subnets then you are going to need some extra network kit for the failover to work ie a GSS possibly as DNS will need to reflect that change almost instantaneously for it to be effective. And once firewalls come into the equation it's get more complicated still as firewalls only deal with IP addresses.

I agree that teaming can create it's own problems and we have had issues with it before. We always setup fault tolerant teaming on our servers as this gives us least problems.

As you say it depends on the particular situation but i still think teaming key servers is the easiest and most cost effective means of redundancy on the server.

Jon