The internet looks very pretty in PowerPoint.
For a business wanting to connect dispersed sites together, it’s a matter of moments to draw up an interconnection diagram with servers in Basingstoke, a customer support LAN in Bangalore and R&D facilities in Boston. In the middle, the placid, cloudy internet, linking everything together at a fraction of the cost of the old days of dedicated lines or today’s MPLS services.
Reality is not so calming. Speed, throughput, latency, reliability are all variable – and costly to get right – through the internet.
Treating it as a dumb pipe is, well, dumb. As a result, a whole field of wide-area network (WAN) optimisation has become intensely interesting: by adding intelligence at the points where your business connects to the internet, a variety of techniques can help speed things up, smooth over variations and get a lot more connectivity for your buck.
The major cost benefits come from delaying the need to upgrade connection speed, and from being able to use lower-cost options when setting up new links, consolidating services or supporting a mobile workforce.
The good news is the techniques can improve performance by factors of 10, WAN optimisation works perfectly well as a software appliance and it can usually be installed without affecting the configuration of your existing connectivity.
Read more about WAN optimisation
The bad news, however, is that every supplier of WAN optimisation does things their own way – useful interoperability between suppliers is rare – and this extends to how they describe their products. WAN acceleration, data acceleration, application acceleration; you’ll find all these terms and more. For example, Cisco calls its version WAAS – wide area application services.
The basic tools of WAN optimisation are compression, deduplication and caching, with more complex networking tricks available such as combining multiple packets, traffic shaping to prioritise important data and route analysis. Also, many optimisers play at a low level with the TCP and IP protocols, increasing window sizes, short-circuiting packet acknowledgements and generally making things less chatty by spoofing or streamlining times when lots of small requests and responses are generated.
None of these are unique to WAN optimisation and if you are familiar with such techniques from elsewhere, you will already know that they are capable of very useful results. Yet, they are also sensitive to different traffic profiles, data types and sizes, and react differently to transmission impairments.
In particular, use of encrypted or heavily pre-compressed data or real-time protocols such as VoIP or video that require consistently low latencies, can need special attention.
You will also need to go beyond the spec sheets if you have to support connections over satellite or into territories with poor or heavily-filtered internet connectivity; optimisation can make the unusable usable - unfortunately, it can also work the other way round.
In general, if you can get WAN optimisation pre-tuned to specific applications or environments, the chances of it working well out of the box are much enhanced. If a lot of traffic is aimed at software as a service (SaaS) providers or other cloud services rather than point-to-point use within the corporate environment, some WAN optimisation support is needed from your ISP or another service elsewhere online.
Most current virtual appliance WAN optimisation products have roots in physical appliances that are, in many cases, still available, but as with so much networking evolution the trend is solidly in the direction of software.
This is also bringing the costs down rapidly and makes it much easier to quickly roll out trials or experimental installations.
The good news is the techniques can improve performance by factors of 10. The bad news, however, is that every supplier of WAN optimisation does things their own way
Here’s a quick overview of a selection of suppliers with different approaches. Note that it is possible to roll your own WAN optimisation through open-source tools, such as WANProxy, squid and Traffic Squeezer, but this is not for the unadventurous or time-poor.
One of the rising stars of WAN optimisation - it calls it acceleration - Silver Peak’s strategy is based around its proprietary VXOA or Virtual Acceleration Open Architecture. This is available to run on Microsoft, VMware, KVM and Xen hypervisors, and combines caching, traffic shaping, deduplication, quality of service, packet amalgamation and re-ordering.
A lot of this happens at the packet level or below, making the system largely agnostic about protocols, traffic types or applications and the virtual approach also lends itself to fitting in with a wide variety of existing installations.
Cisco has not gone down the road of virtual appliances, although it does use the v-word in conjunction with some of its WAAS software that runs on its ISR G2 infrastructure hardware.
Tightly integrated with its own management systems and with specific support for things like Cisco’s video services, WAAS has a big-systems approach to traffic monitoring and context-specific deduplication and the company is also pushing it as a cloud-provider based service.
Cisco also has a mobile-specific variant, which it says is particularly good under conditions of appreciable packet loss and high latency connections often found on wireless and cellular connections.
Another young company, five-year-old Aryaka aims for the mid-market by operating its own global network connected to customers through local points-of-presence. These are accessed through the internet or by dedicated line - it says these offer 20ms or lower latency to 90% of business customers worldwide.
The customer effectively VPNs into the network, which then applies TCP compression, bandwidth scaling and what the company calls optimised access to cloud services, as well as specific support for Microsoft CIFS.
This WAN optimisation ‘as-a-service’ is billed monthly and customers can manage it through a portal, with a promise of 99.99% uptime.
Riverbed combines three standard strands of WAN optimisation - TCP compression, data deduplication and application-specific optimisation - under its Steelhead range of appliances running the company’s RiOS software stack.
Most of these are hardware-based, but Virtual Steelhead runs under VMware ESX or ESXi and Mobile Steelhead runs on Mac or Windows platforms - designed to be paired with an appliance on office premises.
Riverbed has partnered with the Akamai content delivery network in a deal which sees RiOS installed on Akamai’s gateways under the moniker Steelhead Cloud Accelerator. Steelhead users can pay an additional licence fee to activate this, which then adds WAN optimisation between customers and SaaS providers connected to Akamai’s network.