How should a segment my servers into VLAN's in a datacenter design

[jgayou]

I am designing a data center with over 200 servers and I am trying to decide how to segment my data center into VLAN's. I have thought about putting all servers in a different vlan to guard agaist broadcast storms and also a failure that might effect the vlan as a whole. I know I can use storm control to limit broadcast storms.

Another thought is to put all Windows servers into one Vlan and all UNIX into another.

Another thought is to group the servers into VLAN's by the applilcations they support.

Can anyone tell me how they do this in a large data center and if any the pros and cons to doing it that way.

[steven-dunn]

you have to be concerned with how many servers are going to hit their default gateway ? You don't want 200 servers at 100M or more moving data through one router. The simple thing is make as many VLAN's as you have members in the HSRP cluster. Router 1 is active for vlan 100, router 2 is active fore VLAN 200, router 3 is active for VLAN 300, etc. The more VLAN's you have, the safer your servers become from layer two problems, with this, admin gets more complicated.

You also want to concider the uplink connections to your core switches, and use your VLAN and PVST config to loadbalance and layer 3 VLAN over a layer 2 STP trunk. Again you want your active trunk to connect to your active HSRP box to minimise latency. The above will create a VLAN (or two) per uplink so if you have many edge switches, you will end up with many VLAN's which will produce additional STP, HSRP, routing overhead.

You want to keep it simple, but you want to keep it manageble.

[joe-vieira]

For Fault tolerance and load-sharing I'd use 2 Cat4K switches with SUP 3 module (or 2 Cat6K with MFSCs if you can afford it) to route the VLANs. This way it doesn't matter too much if you put all the servers in the same Vlan although you should probably divide the Windows and Unix servers into separate VLANs. IF security is an issue also, you can make these VLANs private.

[davidjmeier]

As stated you're going to need at minimum two 4000s. Since there are 200+ servers in this you may want to go with the 6000s. Obviously you'll need to do layer 3 and have a couple GigE trunks between those two switches to handle any cross switch traffic. Since you have so much riding in this data center you're going to want some horsepower at your gateway. Maybe look into redundant 3700's and use HSRP. Is this a design for a colo type data center or is it all internal servers to your company? I guess if it was the former I would split it out by *NIX / Win, and keep each VLAN down to a /29 and put 4-5 servers in each. This way, if you do have a problem it is segmented relatively well. There are so many variables, it's hard to say with just a few lines to go off of.

Check out the Cisco Reference Designs here:

http://www.cisco.com/warp/public/779/largeent/it/ese/srnd.html

Good luck!!!

[kwftide1]

Are there any rules of thumb when most of the server utilization will come from the WAN? That seems like a recipe for timeouts and retransmissions.

[dixho]

There is an old 80/20 rule. Not sure if the 80/20 rules still used by most network designers. Even since ethernet switches become popular, the rule of thumb is "switch whatever you can. You want to design the network ; so that 80% of the traffic is switched traffic and 20% of the traffic is routed traffic.

[9mmurphy]

The task at hand can be simplified.

I would assume you need to talk to your server group and your applications group. There is a good chance that they will have requirements as to some servers need to be on the same segment as other servers. In the old days, hehehehe... you could group Unix with Unix and seperate the Windows boxes. But today, too many applications are cross platform and my benefit by being on the same segment.

Surely there exists a network where some of these servers are living today. Documenting that infrastructure would most likely expose some dependencies and reveal weak spots. You have the flexibility to build a robust and redundant network. Leverage the opportunity by surveying all interested parties and building an infrastructure that can grow with the business.

Traffic pattern analsys would be very helpful, isolate those chatty devices from the latency sensitive ones and avoid issues down the road.

HTH