Thanks. That's how I feel. However, some people think that a switch somehow increases or propogates the electrical signal between the two boxes,, like a repeater. Thoughts? I think that adding a switch adds an additional and unecessary point of failure. Is there any industry best practice documentation that you could refer me to?

Actually switches do rebuild frames, so you can extend the distance beyond end hosts. But besides adding a point of failure, you also need to insure the switch can actually handle the offered traffic load and also note the switch will add some latency.

We had a server on a 3550-48 XL switch. The interface was rapidly accumulating thousands of underruns causing the cluster to fail over. We moved the connections to another 3550 and it still acumulated the underruns. So we put a cross over cable between the devices and the cluster stopped dropping frames and failing over. The cluster is performing well.

We had another server on a 3550 that was doing the same thing and we put it on a 4507 and it stopped. Where can I find information about the backplane fabric capacities for switches? Thanks for all of your help.

When you are having a problem with the overrruns use the "show controllers utilization " command and see if the port or switch fabric is being overwhelmed. Stats for the 3550 are here. For any switch just do a search on the main cisco page for data sheets, such as 4500 data sheets or 3750 data sheets.

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps646/product_data_sheet09186a00800913d7.html

Summary of various Cisco swithes attached.

Thanks for everyone's help.