Network capacity planning facing data onslaught

Network capacity planning is facing new challenges: enterprises are more reliant on hosted applications while also leveraging wireless capabilities. Some capital expenditures can be offset to service providers, but user service expectations remain high, and new capabilities and technologies are changing the game. “There is a huge growth in Software as a Service , and that’s changing things” says Eileen Haggerty, director of technical marketing for Westford, MA-headquartered NetScout Systems Inc, in conversation with Telemanagement. “Limited resources mean that more things are being outsourced, which means people need new tools to be the eyes and ears of their network.” It’s not news to a network manager that a few applications can hog most of the wide area network (WAN) bandwidth, and that once you move past these top applications the remaining stragglers are inconsequential. The problem is that user behaviour is not as predictable as it used to be, yet application availability and data backup are more important than ever. “We find that organizations are surprised to learn what’s coming across their networks during business hours,” says Haggerty. “Organizations depend on the internet for so much, and with the growth in SaaS we can help them pull a traffic type out of an HTTP black hole, and monitor and track that application.” In practical terms, what this means is that a salesperson using a hosted CRM application can be certain of getting access on a best-efforts basis, because when he or she is on the phone, and needs to pull up customer data, it is crucial that this be given priority. “This doesn’t always require more network spend,” says Mark Tauschek, lead analyst with Info-Tech Research. “What it requires is understanding and prioritization. In fact, prioritization is the biggest issue today. Something like RDP – remote desktop protocol – needs to be prioritized above e-mail and standard file traffic.” This follows the popular wisdom that throwing bandwidth at a situation won’t solve every problem. It makes economic sense because WAN bandwidth can still be expensive, and a software management solution may give you more bang for your buck. As it stands, some network vendors will try to help their customers by selling the minimum required circuit traffic, but this can result in unexpected application degradation and traffic congestion for one simple reason: networks are dynamic, and user behaviour can change abruptly. “There are some problems that come up way downstream,” says Gene Litt, senior director of product management and strategy at Philadelphia, PA-headquartered Shunra Software Ltd. “In these scenarios the application is already on the network, and customers need a WAN acceleration device.” For a company like Shunra, however, WAN acceleration is only one option, and the best strategy for larger enterprises is often to introduce management into the network design process. This is why Shunra also offers WAN simulation technology, which allows network managers to make better, fact-based decisions earlier in the service lifecycle. “This way you can save people money,” says Litt. “We can put applications in a lab setting to see how they perform against various service level agreements (SLAs). We can do this using the standard tools that many enterprises already have, such as load performance. The Shunra box can bring the network behaviour into the lab, then right-size it.” This approach has an inherent and compelling logic to it, particularly if the only other option is to fly blind and pay a fortune laying out more capacity. Canadian organizations have taken note – Shunra’s enterprise customers north of the border include Teradici Corporation, Radialpoint SafeCare Inc, Canada for Commerce Clearing House Inc, Health Canada, ALS Canada, AON Canada, Amex Canada, and Sybase Canada. Enterprise and service provider networks: caught in the middle Companies like NetScout and Shunra are something of a rarity these days – network capacity planning applications are now mostly delivered by the network infrastructure manufacturers themselves. Either way, vendors are finding themselves addressing more and more issues that are beyond an enterprise’s immediate control. “We get caught in the middle a lot,” says Litt. “For example, we might have a large enterprise customer migrating to carrier Ethernet. They need to identify service providers for coverage – then they need SLAs that ensure applications perform adequately for end users.” This is important because carriers don’t only charge for circuits, they also charge for different classes of WAN services such as Multiprotocol Label Switching (MPLS), virtual private networks (VPNs), and Asynchronous Transfer Mode (ATM). “We have a strong presence in the enterprise – always have,” says Haggerty of NetScout. “But over the past four or five years we have been stronger in the service provider market, in particular with wireless.” As enterprises become more dependent on IP-based technology, and as application performance is more closely tied in with revenue generation, it becomes crucial that all performance is assessed. This is becoming increasingly relevant as the 3G footprint has become more prevalent, and more applications are being pushed from the enterprise out onto mobile devices. “We can raise the bar in terms of understanding how services are performing,” says Haggerty. “Enterprises need to know where a problem is coming from, and they need to know the details of what is in a packet. This might mean finding out if there is a mistake in the development of an application so that architecture can be altered.” Big software vendors like Microsoft, as well as in-house developers, need to think about how their applications will work in a service oriented architecture (SOA) environment. SOA is an old term, but the basics apply: many enterprises still have a lot of legacy point-to-point connections, and data can reside in silos, something that has to open up if there is to be an enterprise-wide view. And when that happens, networks have to be ready. “A company like Microsoft will look to the VAR community and network infrastructure vendors like Cisco Systems to help people out,” says Tauschek. “It has to work, or it will reflect badly on the vendors and their partners, whether they are a Microsoft, Avaya, or IBM.” Either way, network managers are now expected to be able to take their concerns further up the stack, and address problems at the application layer. This isn’t just about network availability, then, but falls directly into the need to meet application performance objectives. Increasingly, this is happening in virtualized environments; and fortunately, vendors know that this information needs to be readily available. “There are broad guidelines for things like VDI ,” says Tauschek. “It puts a strain on your network, but vendors like Citrix Systems and VMware will provide recommendations.” Whither bandwidth? High-speed WAN links, fast switches, multi-gigabit transmission – this leaves a lot of IT workers with the sense that, with more bandwidth available, less QoS is required. But it doesn’t matter how advanced the infrastructure is, every network has funnel and bottleneck effects that can negatively affect network performance. “Many people don’t think of bandwidth as a finite resource,” says Tauschek. “It is cheaper now – 10 or 100 megabit Ethernet between sites doesn’t cost much more than 256 frame relay did in the old days – but traffic still has to be prioritized. It’s not the Wild West, where people say ‘We have lots of bandwidth, and we’ll buy more’. That’s just not happening. Instead, they’re finding different ways to limit use.” NetScout has a sniffer product that functions as a firefighter’s tool. Portable, it provides distributed monitoring to assist with intelligent network planning and design. This is particularly relevant in large distributed enterprises – in Canada, Royal Bank is one of NetScout’s customers – and in healthcare. “This is not so much about how much traffic – it’s about what’s in the traffic,” says Haggerty. “We leverage the data that is already in the network, the packets themselves. A vertical like healthcare is huge from a bandwidth perspective, and capacity planning is hard for them, because they don’t have a traditional business day.” This means that QOS needs to take traffic characterization to another level, something that is complicated by the fact that dynamic port use can push around many different applications and drive WAN circuits to over 80% capacity. As a result, congestion can occur in ways that are hard to predict, thus reducing worker productivity. “We identify the web-based, 400/500 errors,” says Haggerty. “The application developers haven’t seen these, and aren’t aware of the potential application degradation. Our customers can develop an ad hoc report, PDF it, e-mail it to the application developer, and say, ‘Look, I am getting twenty second delays in repeated server exchanges: is there any way to cut out some of these handshakes and turns?’” At the end of the day, bandwidth generalizations simply don’t apply for today’s distributed networks. Multiple Extranet and internet POP locations often require a site-by-site analysis. Only that way can bottlenecks – easy to spot, but not so easy to predict – be targeted at the router, hub, switch, or server level.

Network capacity planning facing data onslaught

News | 09/25/2009 5:30 pm EDT

To keep reading, get a free trial.

Already a Subscriber?

Reuse & Permissions

Inside The Wire Report

Information

Hill Times Publishing

Get in Touch With Us