Begin by cleaning up. In many networks, as much as 20 to 50% of traffic is totally unnecessary. In part, the default settings of popular operating systems are responsible for this, because they turn on many unneeded network services. In most cases, users and even administrators are unaware the traffic even exists. The best way to detect this traffic is with a protocol analyzer or "sniffer." Other common culprits are older systems no longer in use. A common example is the NetBEUI and IPX protocols. Not only will removing this traffic free up bandwidth, it will also reduce processor and memory utilization on your PCs and servers.
The next step is to assess your network architecture. Almost anything would do for data, but it's important to remember that QoS will put a load on your network devices that wasn't previously there. A layered architecture like the
Develop a quality of service (QoS) scheme. To a great extent, the best practices here are determined by your Layer 2 technologies. For instance, if your backbone is ATM, you'd use a different strategy than if it were Gigabit Ethernet. Even so, you still need to decide two things:
- How to classify your traffic so that you can distinguish voice traffic from data. Specifically, will you use Layer 2 features, like 802.1Q/p and/or Layer 3 devices like IP Precedence or Diffserv?
- What type of queuing mechanism to use, deterministic or fair? Weighted? Although these are somewhat vendor specific, the best recommendation is one that uses the deterministic Priority Queuing scheme for one queue (voice) and a Weighted Fair Queuing mechanism to prevent the remaining queues (your data) from starving. It's the best of both worlds.
Further, you need to decide details like where you're going to classify the traffic, what you'll use for call admission control (like Resource Reservation Protocol -- RSVP), and whether or not to configure this manually or use some holistic policy management software, or perhaps something in between.
Next, assess your network design. In your design, look for bottlenecks or other points of congestion. If they can't be eliminated, can the voice traffic be routed around them? In the LAN, consider Layer 3 switching. Aside from being orders of magnitude faster than traditional routers, the prices on these devices are getting quite reasonable. This is an especially good move if your older routers don't support your QoS scheme and you have to replace them anyway.
In the WAN, look for ways to minimize delay. For instance, if you're using compression on your WAN links to double your bandwidth, you may be forced to uncompress them to get rid of the delay caused by the compression algorithm. If you're using satellite links, you may be forced to switch to terrestrial. Also consider using the lightest encapsulation method you can find.
Finally, consider the hardware you're using. Making sure your routers and switches support the QoS standards you plan to use seems fairly obvious, but you might want to go a little further. Find as much information as you can about the technology roadmap for the hardware you intend to purchase. For instance, if you're about to upgrade your switches, ask your vendor if the specific platform supports providing power to IP telephone handsets via the Ethernet (CAT 5) cable. Remember that VoIP, while impressive, is still in the early phases. We can expect to see rapid evolution, so make sure your hardware is upgradeable.
About the author:
Thomas Alexander Lancaster IV is a consultant and author with over ten years experience in the networking industry, focused on Internet infrastructure.
For more information:
Read all of Tom Lancaster's VoIP Tips.
Visit our extensive collection of Best Web Links on Voice/data Convergence.
This was first published in October 2003