Sizing a WAN link

[darren-carr]

Hi All,

Just a bit of back ground before I submit my question. We are a relatively small business (<150 users) and currently host our data centre on-site with the business. We host around 130 servers (mix of physical and virtual). We are planning to relocate our business to a professional data centre located around 3Kms from the business. Our carrier has guaranteed 2msec latency in the area we are in. We are planning to deploy a carrier, managed WAN based on a MPLS VPN service. We have three sites that will make up the WAN (main office, prod and DR data centres).

We have spent a significant amount of time reviewing our current LAN, firewalls, etc to try to get an idea of what sort of circuits to deploy in the initial design. We are not seeing a lot of traffic outside of the traffic within the data centre (server to server). With this in mind we have opted for a 100Mb service out of each of the sites. I appreciate you can't tell me if this is or is not enough at day one as you would need a lot more information. I am just looking for some feedback from those who have deployed similar sized networks. What did you guys start with, what issues did you experience, etc.

For our design we plan to use Cisco 2x for the user access layer at the business premises, 3x for the distribution layer, that will then be patched into a Fortgate appliance for firewall policies, before being patched into the carrier WAN that will be made up of Cisco 3845 routers.

In the data centre the 3845 will hand off to a Cisco Nexus 5k which will be the distribution layer, that will have the Nexus 2k devices attached to them.

The 5k will also be attached to a Fortigate appliance that will provide access to the DMZ, Internet, etc from outside.

Does this sound reasonable? How have people sized their WAN links in the past?

Cheers,

Darren

[Marwan ALshawi]

Hi Darren,

for the required bandwidth, practically no one can tel you the exact amount that you need as this depends on the bandwidth utilization from users to the Data center in your network, but generally speaking 100M for a network consist of less than 150 users supposed to be ok, assuming server to server communications dose not go over the WAN

you might us some netflow applications and get some statistical data of peak hours utilization to be more accurate

regarding the Data center switches

" Cisco Nexus 5k which will be the distribution layer, that will have the Nexus 2k devices attached to them."

what you mentioned is N5K extended with N2K ( this is access layer not distribution ) as the N2K is not independent logically only physically it is like N5K with more Ethernet modules so consider this as access from connectivity point of view you might have L3 interface in the N5K but this will be considred routed access layer not distribution layer

HTH

if helpful Rate

[Collin Clark]

As marwanshawi stated, w/o knowing your apps and how they work it is tough, but I agree that 100M is a good start and you should be relatively safe. Once it's up you'll certainly want to monitor so you can find out if you should +/- the circuit. Unless you have a specific need for FCoE I would not use the 2K/5K solution. If you want a data center class switch at the distro, take a look at the 4900M. Check out the 3560X for the access as well. You can't stack them, but they have the same performance as the 3750X and with a significant savings.

[darren-carr]

Hi Marwanshawi/Collin,

Thanks for taking the time to provide some your input. I'm not able to review Netflow statistics as my current switches do not support this. I am using Solarwinds OP Manager to review the performance of my network. The network is currently made up of 4 x access layer switches (2960/3560) that are connected upstream via a Etherchannel to a stack of Cisco 3750 switches that are doing the L3 routing between the VLANS. These switches are then connected to my firewall appliance.

I am specifically monitoring the utilisation of the port channels connected to the stack and rarely do I see double digit traffic up to the stack of switches. I am also monitoring user interfaces and again not seeing a lot of traffic.

In the new design we plan to deploy DFS between the sites so there will be an amount of 'server to server' traffic, but not a lot. Everthing else should be accessed via the WAN from the business premise. We plan to also host a domain controller and print server locally to reduce the load on the WAN.

The reason for the Nexus was to future proof the investment in the network to later support FCoE. It also makes it easier to manage as we just manage the pair of Nexus 5k's as opposed to many switches in a top of rack configuration.

I will upload a network diagram shortly to better support what I am trying to explain.

Thanks again

Darren