15 November 2013

Enterprises face a deluge of data from BYOD, mobility, IOT, and video. To ensure mission-critical information gets through,they need to address WAN and application optimisation as a matter of urgency.

Software defined networking (SDN) is likely to impact IT way beyond the data centre, making WAN optimisation crucial to economic growth. IAN GRANT finds out more.

Software defined networking (SDN) is likely to impact IT way beyond the data centre, making WAN optimisation crucial to economic growth. IAN GRANT finds out more.

The cost of moving a bit of information from A to B may have dropped to a vanishingly small fraction of a penny but it is not zero. Multiply that fraction by the number of bits on the network, currently in the exabyte (1018) per month range, and it quickly becomes a significant chunk of the GDP of the global economy.

At least that’s what management consultancy McKinsey discovered when it investigated the impact of networks on the world economy. Reporting to the e-G8 Forum in 2011, the firm said: “On average, the internet contributes 3.4 per cent to GDP in the 13 countries covered by the research – an amount the size of Spain or Canada in terms of GDP, and growing at a faster rate than that of Brazil.”

But the impact of networking is even more profound. McKinsey found the internet accounted for 21 per cent of GDP growth over the previous five years in the developed countries, a time when most developed economies were struggling to show any growth at all following the banking crisis of 2007/8.

It added that most of the new economic value fell outside of the technology sector, with 75 per cent of the benefits captured by companies in more traditional industries. “The internet is also a catalyst for job creation. Among 4,800 SMEs surveyed, the internet created 2.6 jobs for each one lost to technology-related efficiencies.”

Awe-inspiring as this may be, networking’s contribution to the general welfare is still constrained by several factors. Google, one of the prime movers behind software defined networking (SDN), believes that cost per bit doesn’t decrease with volume. This is because complexity in one-to-one and any-to-any communications requires more advanced forecasting and control mechanisms.

Further, a lack of control and “determinism” (predictability) in distributed protocols means firms have to build for worst case scenarios – i.e. over-provision. Non-standard vendor configuration APIs increase the complexity of automated configurations, and existing routing mechanisms do not allow for scheduling or optimisation to meet explicit objectives such as latency.

While Google was speaking about the internal WANs that connect its data centres, all firms also have to consider the impact of technical network issues related to BYOD, the internet of things (IoT) or machine-tomachine, and staff and customer mobility. They are mostly still working across a three tier public network architecture – core, aggregation and access – which means messages need to be translated at the interfaces, which can add delay and error.
Many have fought this by building their own networks or leasing private lines from network operators, effectively cutting out the aggregation layer. But this is expensive and still requires access to the public networks to support mobility and IoT.

There is also a growing gap between network speeds experienced inside and outside the firewall. While carrier networks are upgrading their core networks to data centre-like velocities of 10/40/100Gbps (and testing 400Gbps), access speeds are small fractions of that.

Relying on the availability of ‘next generation’ all-digital IP/Ethernet-based broadband is also not a realistic option because most carrier implementations are asymmetric, offering faster downloads than uploads. This makes them fundamentally at odds with the growing upload traffic due to cloud storage and services, and video conferencing in particular.

The video threat

The danger that video traffic will congest the networks has been long feared. Joachim Horn, group CTO of Swedish pan-European carrier Tele2, told a London conference in November that 4K video (expected to be commonplace within 12 to 18 months) requires 16x more pixels than HDTV to be delivered at once. This will strain existing networks, particularly access networks.

Since the airwaves are even more constrained by physical limits than fixed cables, much of this traffic will have to go via fixed routes. This swap between what was first broadcast (TV) and what was first delivered point-to-point (voice) was predicted in the 1990s by former MIT Media Lab head Nicholas Negroponte and is now known as the ‘Negroponte Switch’.

To cope, fixed and wireless operators have introduced traffic management or “shaping” policies to ensure the average user gets “an equally bad” service, says telecommunications consultant Martin Geddes. But this results in a service too inconsistent for firms that do business via video conference, archive data at remote sites, or transact business online.

Geddes predicts that networks will evolve into “trading spaces” where services such as video conferencing and email will compete for bandwidth based on their contents’ tolerance of latency – the less tolerant, the higher the price.
This concept is already hard-wired into the basics of WAN optimisation. Already, firms like Exinda allow netadmin staff to create policies that restrict the bandwidth available to different types of traffic and/or users. Setting up and maintaining the policies is a slow, complicated process that can be gamed by corporate politicians.

The shift to cloud-based applications and virtualisation cause further complications. When the basic compute/store/connect infrastructures suddenly become fluid, the complexity of managing the process increases enormously.
As a result, Geddes foresees a time when much of the allocation of bandwidth may be done dynamically under program control, and perhaps by auction. Moreover, this could happen in both private and public networks. Current experiments with TV white space, where unlicensed users can gain temporary access to licensed spectrum on condition that they do not interfere with licensed traffic, are a step in this direction.

Governments presently have no way to monetise this ‘free to air’ spectrum, but they have a vested interest in doing so; and since demand for spectrum is rising, it is almost inevitable that they will.
When network traffic becomes gridlocked, network managers have three options: add bandwidth, accelerate traffic and prioritise mission-critical traffic.

Because of the cost of leased lines, firms have long been aware of the need to maximise bandwidth efficiency. But the business reason for having them in the first place – the user’s experience of interacting with the application – has often been lost in the technical fog of managing bandwidth, re-routing congested traffic, defending against DDoS attacks, and the like.

Don’t ignore the app

Ignoring the application is dangerous to the wealth of a company. A new study of online shopping behaviour commissioned by WAN optimisation specialist Riverbed found that speed is paramount. Around half of internet shoppers in Europe will abandon an online purchase if the web page loads too slowly (UK 46 per cent; France 41 per cent; Germany 51 per cent). For more than one-third, 10 seconds is the cut-off point; after that they abandon the shopping cart (UK 27 per cent; France 42 per cent; and Germany 36 per cent).

Kavitha Mariappan, director of marketing for Riverbed’s Stingray division, says shoppers are now emotionally engaged in their online experience: “Tolerance has evolved into stress, anxiety and, in some cases, anger with slow or poor experiences. This is putting a premium on efficient back office functions, including a website that loads according to the expectations of consumers. These are critical for customer service, engagement and loyalty.”

In its 2013 Magic Quadrant report on WAN optimisation controllers (WOCs) (see graph overleaf), market researcher Gartner said: “The primary function of WOCs is to improve the response times of business-critical applications over WAN links, but they can also help to maximise return on the investment in WAN bandwidth and sometimes avoid the need for costly bandwidth upgrades.”

Most attempts to optimise wide area networking have made things more efficient at Layers 1 to 3 of the OSI network model. But this addresses only half the problem, and increasing numbers of vendors acknowledge they need a more complete answer. But it’s not here yet.

Gartner’s Magic Quadrant for WAN optimisation companies: who will be left after the SDN wars?

WOCs can be either hardware or software. They are usually deployed on the LAN side of WAN routers that connect data centres to remote offices. They ensure mission-critical applications get the bandwidth they need at all times by prioritising their access to network resources and by shaping or restricting other traffic. WOCs can also limit the effects of network latency by optimising transmission protocols and applications. Finally, they can reduce the amount of bandwidth needed by compressing and caching data, and transmitting only changes to them.

CTO and co-founder of cloud operator Exponential-e Adrian Hobbins, says systemic risk often comes not from the WAN but from the application. “IT decision makers need a comprehensive awareness of the applications that run over the company network and a view of how demand for these applications is going to grow over time. Only once the behaviour of all mission-critical applications, including office apps, is undesrtood should a WAN solution be rolled out.”

He goes on to warn against rolling out a WAN all at once: “The delivery of new network solutions can be staged, with stepped upgrades of WANs being rolled out in line with budget and projected growth. This ensures the network is neither overloaded nor under-utilised, both of which hurt profit streams.” Hobbins argues that the better approach is to test applications in both data centre and cloud deployments. This allows firms to see which location is best, app by app.

He believes that this is especially true for SMEs. “The cloud will solve all of the company’s networking problems only if a rollout is accompanied by a robust planning process that ensures the company understands what it is implementing.” Mark Urban, senior director of product marketing for WAN optimisation vendor Blue Coat, says 19 September 2013 was a case in point. “One-third of total internet traffic was taken by smartphone users updating to iOS 7. That was not good news for any IT organisation who had planned patch updates or security fixes that night.” Urban says rather than buy more bandwidth, network managers should look at caching or stream-splitting to boost application performance. The process of caching stores frequently used content locally, reducing the need for WAN capacity. Coupling that with SDN allows to re-configure networks on the fly as demand shifts, he says.

Where’s the end user?

Adding server virtualisation to the issues noted above makes a complex picture even more complicated by orders of magnitude. The present initiatives around SDN and network function virtualisation (NFV) are designed largely to draw a veil over the complexity, leaving network managers with a single smooth and largely automated management interface to their networks.

But if they want to be able to set network policy and ‘fire and forget’, they should look away now. Nathan Pearce, product manager at application delivery specialist F5, says: “We would encourage them to not lose sight of why networks exist; SDN needs complementary services at Layers 4-7 so that organisations can reduce the cost and complexity of deploying software defined application services.”

While most network managers have governed events at the Layer 3 and below, they are increasingly required to address matters at Layers 4-7. This is because managing the customer experience, as indicated by Riverbed’s research, is now the competitive edge for attracting and keeping both customers and staff.

However, apps and customer experience are largely unfamiliar territory for network staff. But can they afford to ignore this? After all, the hype around SDN and NFV largely disregards the customer experience aspects of networking. It’s unlikely. The most recent and high profile sign came in November from Cisco. The market-leading vendor made its new focus on applications explicit with the launch of Application Centric Infrastructure (ACI). The brainchild of Insieme, the systems development house it largely financed and is now acquiring for $863m, ACI seeks to extend SDNlike programmability from a central controller to all seven network layers using ‘open’ APIs.

Jim DeHaven, head of data centre and virtualisation in Cisco’s UK and Ireland office, says most SDN implementations are overlays that concentrate the intelligence in the controllers so that all network staff do is connect pipes and IP addresses. “We’re making the networking team far more relevant in the world of SDN. Cisco is leveraging the skills they’ve developed over the last 20-25 years around troubleshooting, scaling, development of data centres and investment in Cisco technology.”

DeHaven reckons ACI gives enterprises “extremely powerful” options for both private and hybrid or federated cloud implementations. It will also allow host providers “to build their own AWS-like (Amazon Web Services) services,” he adds.

Key to ACI is the Application Policy Infrastructure Controller (APIC). According to David Krozier, telecoms infrastructure analyst for market researcher Ovum, this supports a common policy framework that Cisco will extend to bring compute, storage, and network infrastructure under a “common pane of glass”. He notes that while ACI generally supports SDN protocols such as OpenFlow with merchant silicon, to get extra features you’ll need Cisco ASICs. “While this may raise the hackles of those who believe future networks should be based on generic hardware platforms, this approach is unlikely to match the performance capabilities of ACI.”

The new 100Gbps-ready Nexus 9000 series switches in the ACI run a new optimised version of Cisco’s NX-OS and support up to 1.92Tbps per slot. The eight-slot Nexus 9508 is available now. The APIC, which allow applications to be untied from an IP address, and the new NX-OS will be available in April 2014.

Cisco claims that the APIC will cut application deployment times to minutes, while application network profiles and L4-7 network service automation will “tear down” apps that are no longer needed, conserving bandwidth.

“Only once the behaviour of all mission-critical applications is understood, including office apps, should a WAN solution be rolled out.” Adrian Hobbins, CTO and co-founder, Exponential-e

HP’s response was scathing. Nick Watson, EMEA VP of the firm’s Networking division, says Cisco is ignoring the SDN movement “once again”. He accuses it of continuing to create a “hardware-defined” alternative that locks customers into Cisco, thus denying customers “the economic and game-changing simplification, automation and application development benefits promised by SDN”.

Cisco is also battling VMware, which bought SDN specialist Nicira last year and introduced its NSX network virtualisation platform in October. In a blogpost in response to NSX, Cisco CTO Padmasree Warrior said: “A softwareonly approach doesn’t scale and it fails to provide full real-time visibility of both physical and virtual infrastructure. (It) does not provide key capabilities such as multi-hypervisor support, integrated security, systems point-of-view or end-toend telemetry for application placement and troubleshooting.” That means companies have to tie together multiple third-party components, adding cost and complexity, and manage them. And while WOC makers might have to pick winners, for now, while Cisco and the world war over SDN, network managers need to avoid the crossfire. But service providers with bandwidth to sell should be rubbing their hands in anticipation.