10 vs 100 eth for end users vs auto for server throughput

[vipperdigger]

There is a lot of debate here for setting up users with Auto/Auto vs hard coding 10 full. One group says hard code both end at 10/full as not to overwhelm our servers (100 Meg) , others say 100 Meg for end users will get data out quicker so therefore our current day servers should be able to keep up. Cisco says set both for auto/auto but fails to state that 5000 series switches don't do auto/auto very well (4000, 4500 and 6500 are fine it appears). Any links to where studies have been done on a large lan networks and showing where all users are set at 10/full then changed to auto and the effects on through put. We have 1500 users accessing a server farm. Users connected to 16 4006/5509. These switches are connected to MSFC cores via Gig fiber. The server switches are 2 4006 connected to the MSFC via Gig fiber. We have set up the VLAN's for breaking out divisions/floors/servers etc....
Sooooo if users are accessing the switch at 10, uplink port at 1000, to servers at 100, and we now set users to 100, will we bottleneck at the servers, ya know, the old having 10 10 lane highways merging into a 100 lane highway is OK, but having 10 100 lane highways merging into a 100 lane highway, well then we have LA at rush hour. Any thought, and links, any advise.

 

[baddos]

You would have to look at the server's network utilization to know for sure. If that's a problem though, you can pickup 1gb nic cards cheap, and if your really wanted to get some bandwidth go with two intel 1gb nics teamed.

I would probably have more than one 1gb link between the chassis switches and your core though. If you have a high port density in those switches, you should create a 2gb or better etherchannel port group.