Get packing not purchasing

Why throw bandwidth at your network when you can save money by using it more effectively? Danny Bradbury explains how to push...

Why throw bandwidth at your network when you can save money by using it more effectively? Danny Bradbury explains how to push your infrastructure further

How many elephants can you fit into a Mini? That's easy - the answer's four: two in the front and two in the back. It may not be true in real life, but that's because your average elephant takes up a finite amount of physical space. Network packets, on the other hand, take up no space at all, so you should theoretically be able to fit lots of them down a small pipe. The trick is to do it without slowing your end-users down.

Most people don't bother with the equation. It is easier to buy more bandwidth, and this has become a standard technique for many network managers as the level of network traffic increases owing to data-heavy office suites, IP telephony, video and other applications.

It isn't the most elegant solution, however, and even though bandwidth is relatively inexpensive these days, it will still show up on your balance sheet. Ideally, taking more intelligent steps will make the most efficient use of your corporate network infrastructure. Capacity planning is one of the most useful techniques available to network managers and possibly one of the least used, because the methods involved are not well understood. Calculating the baseline is a particularly critical part of any capacity planning process, but do you know how to go about it?

Peter Ellens, networking consultant at IT services company Steria, says network managers must periodically monitor their traffic at various times, changing the frequency with which traffic is monitored, to identifying as many of the peaks and troughs as possible. "No two weeks will be the same. If you happen to choose the month when everyone's on school holidays, then it might give you a false reading," Ellens explains. "It's an ongoing thing: you should set up the baseline and monitor it for a week, then perhaps the following month, monitor it for a different week."

The volatile nature of business may lead to unpredictable network conditions at various times, and so it is very difficult to get a perfect network baseline that you can use to predict network capacity with complete accuracy. Nevertheless, producing some sort of chart showing network traffic over a set period of time will at least give you some idea of how demands on the network could vary throughout the weeks and months. Ellens suggests monitoring it for one week per month, for three months. Make sure that you pick a different week in the month each time, and if there are particular times when you know that demand on the network will peak, then factor those into your baseline as well.

Armed with this basic metric, you can then talk to your business staff, and find out what sorts of application data they expect to be passing across the network, and how often. Find out how much traffic each PC in a given department puts out on average, for example, and then grill the line manager about how many more PCs he or she expects to add in the next year. Information like this will help you to plan the capacity of your network intelligently.

Knowing how much information passes over your network (and ideally its source) is one step on the road to intelligent network usage, but there are other, smarter things that you can do. Traffic prioritisation is an especially useful technique for companies finding their networks bogged down with traffic who don't simply want to add more bandwidth in a bid to temporarily solve the problem.

Traffic can be prioritised in various ways, depending on the requirements of your business. If you have a coherent directory system, for example, you can filter traffic on a user-by-user basis. Another option is to prioritise according to the originating port, making it possible to set rules for traffic from particular departments, perhaps.

The problem is that most people don't really understand the type of traffic flowing across the network, explains Mark Urban, product marketing manager at Packeteer, a company selling traffic analysis and prioritisation hardware. "Generally, the intelligence levels that users are dealing with are at the IP routing addresses and the port numbers that the IP is using to communicate with the application," he says. "That doesn't give you enough information to understand what's running on the network. We analyse Layer 1 through to Layer 7 to discern HTTP, Napster and Oracle traffic, for example."

When discussing layers 1-7, Urban is talking about the Open Systems Interconnection model stack which these days has been replaced in large part by the Transmission Control Protocol stack in the commercial world. Port-based prioritisation exists at lower levels of the stack, and the general rule is that the further up the stack you go, working through port-based packet analysis at the data-link layer and then hitting the higher application layers, the more flexibility you will have to slice and dice information about the types of traffic passing over your network.

When you've analysed the type of traffic passing across your network according to the criteria you've set, you'll be able to set prioritisation policies in place. There are various technologies available for this purpose. IEEE802.1p, for example, lets you tag packets in a bridged Lan environment for faster transfer. The Internet Engineering Task Force's Multi-Protocol Label Switching (MPLS) is another technology which replaces traditional Layer 3 routing tables with its own predefined routing path. The label attached to a packet can also contain network prioritisation information. Then there is Differentiated Services, another taskforce standard that uses a packet field to mark packets with routing rules. This is mainly designed for one-way, asymmetric traffic flow.

With these technologies, you will be able to condition network usage so that the important applications or users stay in the fast lane. Jokers who send a 5Mbyte attachment to everyone on their internal e-mail lists, or who decide to play Quake with their friends on the network during the end-of-year reporting period, suddenly won't be able to slow down traffic from the accounts department while the bean-counters are trying to prepare figures for the auditors by your company's reporting deadline. Similarly, users of Napster-style file sharing services won't be able to bring the network to its knees by downloading the latest Britney Spears single.

Traffic prioritisation can relegate the users of non-business-focused services to the backwaters of your network, but if you've configured your network properly, you should be blocking access to file-sharing sites such as Audio Galaxy and Limewire anyway. The real bonus of prioritisation will come in regulating internal network traffic, rather than illicit packets from outside. But again, you should be able to reduce a lot of the less relevant internal network traffic by educating end-users or, better still, configuring their PCs to be network-friendly. For example, stopping if you teach end-users to not append copies of e-mails to their replies, lengthy dialogues will not cause the amount of information sent in e-mails to grow exponentially.

Similarly, watch out for end-users who send large audio and video files. Even the most sane employees seem capable of sending bulky joke material across the network. If you're providing Microsoft Outlook over the network, for example, there's an option in the application to turn off e-mail appendages. Use it, and watch your network traffic go down.

When you move from the Lan to the wide area network (Wan), it becomes more important than ever to use your bandwidth effectively, thanks to the high cost of data throughput in this space. There are options for companies here. Traditional Frame Relay and wide area asynchronous transfer mode links are fast being superceded by alternatives. Companies that don't mind using the Internet are starting to use Digital Subscriber Line (DSL) to communicate between offices, for example. Even if you are worried about security, there are many virtual private network options to help you keep your data secure when running it over a DSL link.

Kieran Fitzgerald, Wan manager at packaging solutions company Smurfit, is planning on moving the majority of his 30 UK sites away from a BT-operated Switched Multimegabit Data Services (SMDS) network to a DSL local loop infrastructure. "In each country we would have a star topology for our network and they'd be connected by ADSL into our central site," explains Fitzgerald. "SMDS is very expensive compared to Frame Relay and ADSL. We'll just replace the routers, put in new circuits and move across. We won't even move through on a trial basis. It'll reduce the cost by almost e300,000 [£191,000]."

The changeover, which is being facilitated by network service provider Vanco, will save about 60% on Smurfit's wide area network.

Other companies also offer cost savings on Wan bandwidth. Network service provider Packetexchange, for example, offers an MPLS-based network which looks like Ethernet to end-users and which, according to staff, has a pricing structure that increases in small amounts of just 1megabit per second.

The bottom line is that on your network, knowledge is power. Understanding traffic flow will help you control it. While this may necessitate some investment in decent network monitoring tools, it'll enable you to run a tight ship in the long term, and in the current climate, that can only be a good thing.

Top 10 tips for smart network usage

1. Configure your subnets correctly. Avoid overloading a network segment with too many data-hungry workstations. Divide them appropriately. Spend some money on workgroup switches to provide dedicated bandwidth to your workstations

2. Educate your end-users. Warn them about things like sending huge, unnecessary attachments

3. Erect a firewall between your network and file-sharing sites, and other bandwidth-hungry, non-business-related distractions

4. Monitor your network traffic for one week per month over three months, choosing a different week in each month. Use the information to build a baseline, and use it as a capacity planning resource

5. Talk to your end-users. What applications are they planning to use? How many staff will be joining in the next 12 months?

6. Analyse the types of traffic running over your network and map it to your business processes. Find out which is more critical, and prioritise it

7. Check your network interface cards and other networking equipment regularly. One noisy network interface card can bring down a whole network segment, and even a whole network

8. Consider moving to Digital Subscriber Line-based virtual private networks for branch offices. Weigh up the costs of more traditional ISDN/Frame Relay/dial-up links, but be sure to give yourself enough bandwidth

9. Work with network service providers to get just the level of bandwidth you need. There's no point paying more for bandwidth you don't need just because the provider you're working with won't sell you exactly the data transfer speed that you want

10. Buy a decent network management tool. Something like HP Openview comes highly recommended. You must know how your network is functioning daily if you are to carry out trends analysis.

Read more on Voice networking and VoIP