Disaster recovery requirements are now far more pressing and diverse than they have ever been, due to a combination of compliance-related regulations, data centre expansion, the proliferation of remote offices and employees, and server virtualisation. That means you need to be able to exchange data between primary and recovery sites as quickly as possible with as much bandwidth as you can afford between the two endpoints. In many cases, you will need to make the most of the bandwidth you have, and one way to do that is to deploy WAN optimisation technology.
This additional complexity in terms of how and where disaster recovery is deployed creates potential bottlenecks, especially when running across public networks. Latency in particular becomes a major problem on relatively low-bandwidth connections with multiple hops between the source and destination points. WAN optimisation deals with these issues.
By minimising latency, making greater use of the available bandwidth and potentially reducing the amount of bandwidth required in the first place, WAN optimisation can help reduce disaster recovery costs, which can make disaster recovery feasible over a link where it previously wasn’t possible.
WAN optimisation methods
WAN optimisation products typically are not based around a single acceleration or optimisation methodology but rather a combination of techniques, including the following:
- Bandwidth management and traffic shaping. This involves assigning a priority to a particular type of application, which has an effect on the order the traffic is sent in and in the amount of bandwidth the application is guaranteed regardless of other traffic on the network.
- Protocol optimisation. This takes protocols that are inefficient over the WAN and makes them more efficient, typically by converting a time-consuming serial communication process into a more efficient parallel process with many communication tasks handled simultaneously.
- Byte caching. This observes repetitive patterns in all TCP-based application traffic and symbolises those patterns with a token that replaces that traffic, thereby minimising data transfer.
- Object caching. This is protocol- and application-specific. If the cache contains the object, the user is immediately served the object from a local store, virtually eliminating latency and WAN bandwidth consumption. If the cache does not contain the object (or contains an outdated version of the object), for that particular transaction, a new object must be reloaded into cache and the performance gains are realised in full the next time the object is requested. Some forms of object caching compare new with old and only send the changed data and reconstruct the object locally.
WAN optimisation vendors and products
The WAN optimisation market has expanded over the past five years, with Riverbed Technology and Cisco Systems leading the way in terms of market share, according to Gartner, but with several new and innovative companies providing a different angle on the technology. From a disaster recovery perspective, the WAN optimisation players specialising in this space are those with high-end products able to support higher bandwidths.
Riverbed’s Steelhead appliances provide end-to-end optimisation using a combination of the techniques outlined above in what the company calls an "application agnostic algorithm" designed to maximise optimisation of each application and achieve large-scale data deduplication. A typical example of a Riverbed device for disaster recovery applications is the Steelhead 7050, which is designed to support as many as 100,000 optimised TCP connections and gigabit-per-second WAN capacity.
Cisco’s WAN optimisation products are centered on its Wide Area Application Services (WAAS) offerings, which use application/protocol-specific acceleration techniques validated by the application vendors. WAAS uses standards-based compression and context-aware Data Redundancy Elimination (a Cisco extension of the standard methodologies), plus TCP optimisation techniques in combination to reduce bandwidth consumption and improve application throughput. An example of a WAAS offering is the Cisco Wide Area Virtualization Engine (WAVE) 8541, which offers up to 2 Gbps throughput and 150,000 TCP connections. It is supplied as a standalone appliance or as a module for the company’s Integrated Services Router (ISR) range. Cisco’s virtual WAAS (vWAAS) can also be integrated with its Nexus switch range.
In direct competition with the Steelhead and WAAS appliances is Blue Coat Systems’ Mach5 technology. This comes in appliance form, and the range is topped off by the SG9000-40, with gigabit-per-second performance and support for 100,000 TCP connections. Mach5 makes extensive use of object caching, meaning it is well suited to accelerate applications such as video.
Silver Peak Systems’ NX, VX and VRX appliances focus on TCP acceleration, using a combination of TCP window scaling and what it calls "high speed TCP." The former extends standard 64 KB TCP window size limitations to as high as 1 GB, while there is also a modification to TCP’s congestion control mechanism, which changes how the TCP window opens and closes during data round-trip events.
F5 Networks’ Big-IP WAN Optimisation Manager is available as an add-on module for the F5 Big-IP Local Traffic Manager device or as a standalone hardware or virtual appliance. It focuses on advanced compression, encryption and traffic control technologies to reduce data replication times and improve bandwidth usage.
Talon Data Systems and BitSpeed focus on maximising bandwidth across gigabit to 10 gigabit connections. Talon Data's NetFlight range of appliances use a proprietary WAN protocol plus optimisation of the storage disk transfer process for read and write acceleration and can provide sustained performance at 650 megabytes per second (MBps), according to our testing. BitSpeed's Velocity software sits in a Linux-based appliance and is effectively a rewrite of TCP that uses multiple, parallel streams to support transfers at up to 10 Gbps.
Juniper Networks also has a range of WAN optimisation appliances, notably the WXC platform, which supports links as fast as 45 Mbps links in the WXC590 model. The WXC provides TCP acceleration, SSL optimisation and what Juniper calls AppFlow acceleration: application- and protocol-specific optimisation methodologies designed to accelerate each application/protocol combination to its peak.
What’s next in WAN optimisation?
One innovation just around the corner is the ability to predict data transfer requirements and begin the transfer process before it has even been requested, based on analysing user and file behaviour remotely. So, in a scenario where data has to be retrieved urgently, the request will already be primed, resulting in a much faster retrieval time. There is also an argument being put forward to replace Multiprotocol Label Switching (MPLS) networks completely with simply architected networks using WAN optimisation. MPLS is a notoriously expensive methodology for deploying and managing networks, so the rationale, from a disaster recovery perspective, of simplifying the network architecture and, at the same time, optimising what bandwidth is available makes a lot of sense.
Steve Broadhead is founder and director of network equipment test lab Broadband-Testing and contributes regularly to ComputerWeekly and other IT publications. He is also a regular speaker at global networking events such as Netevents.