1

I have a 50Mbs Satellite link and a 10Mbs Microwave link supplying a very remote location.

Behind these links, I have a 6,400 seat network - with about 3,000 signed in at any one time.

My goal is to send all of the Voip traffic (Google Chat, Magic Jack, Skype, Speakeasy, Vonage, Vonage PC, Yahoo) through the microwave link which has 100ms latency. The rest of the traffic can utilize any remaining bandwidth of the microwave link with excess being diverted to the higher latency (600ms) satellite connection.

The problem I've had so far is that most automatic routing configurations weigh the bandwidth heavily for preference - and I'm only wanting latency considered.

Additionally, I don't know if this can even be handled with the routing hardware I have at my disposal (Cisco 3640, 3745, & 3845).

Any recommendations (or really good starting points) would be greatly appreciated.

2 Answers2

1

You are going to need qos and traffic classication.

will do every thing you need. (and more)

Make sure you two boxes in for fail over.

I'd also avoid or rate limit skype (if possible). It has the habit of consuming all resources, it's a bandwidth leeching program!

You can implement qos on your cisco',s but you'll find pfsense and easier way to do this.

You'll also get add bonus of a quality firewall and be able to rate limit and traffic shape you clients.

Make sure you use quite beefy machines pfsense. or add more gateway to spread the load.

Get yourself a pfsense support contract, and donate to the project. It's awesome.

The Unix Janitor
  • 2,458
  • 15
  • 13
  • -1 ???????????????????????????????????????? – The Unix Janitor Mar 23 '10 at 21:12
  • 2
    Your answer looks more like an advertisement for pfsense than an actual answer. If you add in how, at least conceptually, you would configure pfsense for this, I suspect you'll get some upvotes. – Vatine Mar 24 '10 at 15:31
1

If you have an easy way of classifying the voice traffic (auto-detecting RTP would be one way), you should be able to use policy-based routing to do this. I don't have an example handy, unfortunately (it's been a few years since I needed to do PBR).

On a Cisco, you stick a PBR as an incoming policy on the ingress interface(s), then use an extended ACL to match for what you need (it MAY be possible to use a class-map for this, I haven't tried, I only needed to route on source addresses (load-balancing through in-line HTTP-mangling devices to optimize the HTML for mobile devices, more expected concurrent users than each inspector box could handle)), then set next hop. If a packet hasn't been routed by the PBR, it's handled by normal routing. If you try to set an unreachable next-hop, the processing falls through to the next rule in the PBR.

Ideally, I'd want some dynamic routing protocol in place across the microwave link and use a loopback on the remote end (announced via said routing protocol) as the next-hop address, so any link failure ends up with the next-hop being unreachable fairly rapidly.

As to what routing protocol to run oiver the link, sorry, I don't know what the best choice would be. I'd probably start by using OSPF or whatever you're comfortable with (although, since you're probably looking for a quick failover, RIP is probably not the right choice).

Vatine
  • 5,440
  • 25
  • 24