LAN switch configuration

dpropson · ‎04-27-2007

I work in a CAD-intensive network with Cisco Catalyst 3560G switches. We have a rack of 5 switches that feed a floor of users and all the servers and all the feeds from the other 3560s on the 3 other floors. The 3560 are all GB and I have CAT6 connecting switch to switch.

How do I know that my setup is optimum? Should I spread the load of the servers to several different switches? Should I bring all the switches from the various floors to one switch or divide them among several? Any thoughts are appreciated.

griffijo · ‎04-27-2007

I like to limit the diameter of the network as much as possible, meaning if you can bring all your switches directly back to an aggregate switch then do so. In addition to minimizing the uplinks you can also make that switch your spanning-tree root which also helps with troubleshooting. With respect to spreading the servers that is a good question. I used to think that spreading the servers across switches, or placing the server closer to the user was a good idea, but over the years I changed my thinking with that. I'd rather collapse those servers onto a single swith backplane for the sake of management, security, etc. Is this one vlan or do you have multiple vlans? Depending on the size of the network and the number of servers you might consider something like a catalyst 4500 at the "core" of the network.

Iain · ‎04-27-2007

If you're looking for maximum performance you may also want to consider the new(er) 3750-E switches (10GB uplink capable). These have the Stackwise technology which means a large stack of these switches can be combined with high speed 32GB inter-connects to form a single logical switch.

You should consider configuring etherchannel on your uplinks and server connections.

I think something which is often overlooked is monitoring. If you aren't able to monitor your performance you have no idea where the bottlenecks are!

For example: it doesn't make sense to add four NICs to your server if the disk or processor isn't capable of utilizing them. I think the best scenario is when you can monitor at all key points.

There are lots of software packages that can help you do this. Cacti is an open source package that can do it for free.

http://cacti.net/

If you have a SAN, you will want to monitor performance from that end as well. (i.e. if you have a Xiotech SAN, they have a product called Storage Monitor).

If you want to get more granular with your network monitoring, you might consider implementing a NetFlow collector.

http://www.crannog-software.com/index.php?go=Product.ShowDetail&ProductID=1

Once you start collecting data, you may be surprised to find that your weakness is in an area you never would have considered. Then you can make more intelligent configuration changes.

HTH pls rate helpful posts

griffijo · ‎04-28-2007

I like the 4500 over the 3750 stack as a distribution layer switch. It's a little more scalable and supports some features, like Netflow, that the 3750 does not.

Iain · ‎04-30-2007

Agreed. However, if the $$ aren't available for the 4500 or 6500 series, the 3750 is a nice switch.

We do our Netflow export from the core. So the Netflow feature isn't important for us at the access/dist layer. Of course, we aren't doing L3 access layer .. yet. SNMP (MRTG) where performance monitoring of access/dist is required.

vanguard1 · ‎04-30-2007

Any thoughts on using a 4500 switch at the core on a 700 user network.

letmeinnn · ‎05-04-2007

ia m using 3750g switches for the core on a 800 user network.

utilization is decently low

wish i could spend the $$$ on a 6509

Cory