When you're mapping out your WAN, it's not quite a case of "Here be Dragons", as mediaeval cartographers used to write, but there's definitely still something of a Wild West feel out on the edges.
Not only can that wildness be a bit disorientating for anyone used to the visibility and manageability of a LAN, but it's not so good for your applications either. But short of making all your application developers work at the end of a 64k leased line so they understand what a WAN will do to how their software works, what can you do?
It's this that has spurred the development of WAN optimisation and acceleration appliances - boxes that either sit in the data centre and optimise how web transactions are handled, or sit on each end of a WAN link and pre-process network traffic to add compression, reduce latency and cache shared content.
And thanks to trends such as teleworking and server consolidation, there's no sign that the need for these technologies will go away any time soon.
"Especially in the IT industry, people have been there a long time, they want the ability to work from home," says Scott Safe, marketing and business development VP at Network Physics, which builds network traffic monitoring devices.
"There is still a belief that server consolidation will reduce cost," he adds. "Where the mantra was decentralisation, so you put the servers near the people, the cost got out of control because WANs were too expensive."
"The pendulum of centralised versus distributed swings back and forth every 10 to 15 years," reckons Craig Stouffer, world-wide marketing VP at WAN acceleration developer Silver Peak Systems. "Perhaps in 10 to 15 years, once SOA is fully implemented, it might become distributed again. Today you can't really implement full peer-to-peer SOA because the bandwidth isn't there."
He adds, "A huge trend we've seen is the globalisation of enterprise applications. An organisation of 500-plus employees will typically have over 100 applications on the WAN, and the majority of those are proprietary, business-critical, own-developed applications that are now being distributed across the enterprise."
According to Safe, it's not just a matter of latency and inadequate bandwidth though. "The next problems are errors from congested or oversubscribed points on the network," he says. "TCP doesn't do well when packets start falling on the floor. Ten years ago the problem would have been lossy lines, now it's more likely to be congestion, but the result is the same - dropped packets."
The moving bottleneck
The next problem, as many who work in the application acceleration market will admit, is that clearing one set of bottlenecks is quite likely to create - or bring to light - bottlenecks in other areas. WAN optimisation can even break networks, according to Mike Banic, senior director of Juniper Networks' application products group, which recently added new models to its WX (ex-Peribit) and DX (ex-Redline) optimisation families.
"We have a customer with WXs on 140 remote sites, he says it 'breaks networks' because it's like putting too powerful an engine in a car - it causes stress failures to occur earlier than they otherwise would have done," he says.
He adds that this is one reason for not integrating the optimisation capability into other WAN equipment such as routers or UTM devices just yet: "As a dedicated appliance, it's a lot easier to adjust when you find those stress cracks."
The bottleneck can also move to the application or the server, notes Craig Stouffer - and even if they don't break under the strain, other problems can arise.
"Typically, the application is network-limited before we go in," he says. "We make the WAN act like the LAN, so the number of server transactions goes up by an order of magnitude and either the server falls over, or they only get 10 percent improvement in application performance because now it's server-limited."
Baselining is key
Safe adds that if you want to make an application run well over a network, you have to understand what's there already, not least because you can't really optimise unless you know what you're optimising.
"The biggest problem is people don't know what running on the network and what the traffic profile is. We've thrown everything onto the network in the last 10 years, and now organisations are struggling with how to control that," he says.
"The application and network teams don't speak the same language, but we are seeing a big change as they realise that changes to an application can have a huge impact on how that application runs over the network."
"There's so much being thrown at enterprises now, they're in fire-fighting mode. Fortunately, with server consolidation we're starting to see organisations doing baselines to see how they can accommodate the shift to new architectures."
The evolution of standards for measuring application performance is making all that baselining easier, claims WildPackets's European sales engineer Chris Bell - WildPackets now includes Apdex scoring in its OmniPeek network analysis software.
"A lot more sense is prevailing now," he says. "Having standard metrics, such as Apdex, is going to help people get their heads around what's happening on the infrastructure - the most important thing is establishing a baseline. You always find things in the last place you look, so you need to look there first. That's what metrics help you do.
"Once people start to understand what's happening in their systems - all their systems - it breaks down barriers. But there will always be a bottleneck somewhere, something that can be improved - it's a continual process of improvement."
Get your retaliation in early
Bell adds that the key for network managers is not to wait until things go wrong before doing something to make them better. He argues that a process of continual improvement can mitigate problems and bottlenecks.
"If you come into work, and email is slow, you accept that," he says. "But there's a huge amount of efficiency to be obtained by taking a proactive approach and setting yourself targets for improvement. For example, the next thing that organisations need to worry about is mobilising applications."
For many organisations though, simply learning what WAN optimisation can offer - as many did by coming to Techworld's recent conference on the topic - will be a big step forward.
At that conference, a spirit of 'co-opetition' became visible almost for the first time, as companies realised that they need to stop competing over those network managers who've already installed WAN optimisation technology.
"The big problem is awareness - there's a whole bunch of people who haven't even heard of this stuff yet," Stouffer admits. "It's estimated that there's four million branch offices world-wide, and that number's growing four percent a year. The world-wide installed base of WAN optimisation products in 10 years is about 60,000 units - and that's including the whole Packeteer user-base" - although the company has struggled to keep up of late, its bandwidth-shapers pretty much invented this whole market.
What will convince the laggards? We won't know that until we see it, predicts Chris Bell: "It's at the stage of mobile phones 10 years ago - we're waiting for an application like SMS to come along."
Keep your eyes peeled therefore - it could be peer-to-peer SOA, or it could be mash-ups, or it could simply be growing mobility among your users. And whichever it is, don't let yourself get caught out.