"WAN optimisation started out as a repairman model. We need to stop fixing problems after they occur and start building WANs that automatically optimise themselves." So argues Dick Pierce, the CEO of Orbital Data, a relatively recent entrant into this market which has eschewed the application-specific optimisation modules used by its rivals in favour of a self-tuning technology which Pierce says makes the WAN application-aware - or as Gartner calls it, application-fluent.
Orbital recently jumped from being a specialist TCP flow optimiser with a number of satisfied multi-site customers, incuding Sony Pictures, document digitiser Peelle Technologies and seismic analysts Petroleum Geo-Services, to offering a spread of WAN acceleration technologies comparable with the likes of Expand Networks, Juniper-owned Peribit and Riverbed Technology.
"We started up in 2002, we've been selling product for around 18 months," Pierce says. "We would mark the Packeteer days as the beginning of WAN optimisation. Since then a whole host of companies has entered the market, each one bringing something new."
What Packeteer did in the late 1990s was build a business around rate-shaping. For pretty much the first time, it allowed an administrator to allocate WAN bandwidth by application or protocol, for example to prevent bandwidth-greedy traffic such as HTML from swamping more critical traffic such as SAP or Oracle.
"Packeteer does data queuing by flow, it's the equivalent of Layer 4 QoS - it filled a huge gap left by the relatively crude Layer 3 QoS on Cisco routers. But rate shaping was nothing to do with protocol optimisation, it was simply an effective queuing scheme," Pierce says.
So WAN acceleration developers came up with techniques to optimise network and application protocols by minimising their chatiness, for example, and they added more techniques too, such as data compression, and now disk-based file caching. Others focused on areas such as load balancing, SSL offload and transaction security.
WOCs, WAFS and ADCs
The result has been the emergence in recent years of three basic categories of accelerator: application delivery controllers (ADCs) such as F5, Fineground (now owned by Cisco), Netscalar and Zeus, WAN optimisation controllers (WOCs) like Allot, Peribit, Expand and Packeteer, and wide-area file systems (WAFS), for instance DiskSites and Tacit. These categories are now converging, especially the latter two.
Pierce claims each of the acceleration techniques has disadvantages as well as benefits though - and of course, not all of them are appropriate for every application. This, he says, is why Orbital has come up with its AutoOptimizer engine which dynamically works out which to apply to each data flow.
"Peribit and Expand tunnel through the WAN, so they can't see the packets, while compressed traffic breaks network monitoring," he says. "Each technique has its pluses and minuses, each has scalability disadvantages. Everyone is catching up by adding new technologies, but everyone gets too wrapped up in the repair argument.
"We use all the techniques but auto-optimise - we have built a flow-control engine as the base, it sits transparently in the network and nothing has to change. All vendors are application-transparent, but not all are network-transparent - we are the only vendor rebuilding the TCP datagram, for instance. It's fully TCP-compliant, using spare fields in the header.
"Adaptive optimisation looks at the network level first for flow control and rate-matching. The second level is looking at the data itself, most importantly is it compressible, and which protocol is it? The third piece is the application layer and being able to distinguish VoIP from FTP or SAP, say, because each application has different needs for best performance."
Jack of all trades, master of none?
Other WOC vendors have criticised Orbital's approach, arguing that only by tuning for each major application will you get the best results - and that means doing application-specific software modules. However, Pierce believes that "a balanced mix, a painless default," makes more sense in the long run.
"We don't specifically deal with MAPI, for example, but it still gets benefit," he says. "If you deal with MAPI specifically you risk Microsoft tweaking the protocol and making your work redundant."
The big thing about AutoOptimizer as far as he's concerned is that it lets the WAN adjust itself to suit the apps running over it - and that in turn should mean it needs less management. "Our box also auto-discovers, all it needs is an IP address. The goal is to install it in less than 20 minutes," he adds.
"The unifying factor in the convergence of WAN optimisation is management - tying it all together," he continues. "We've been working very hard to integrate additional components in a way that maintains simplicity - the industry has grown up with lots of knobs and dials, but it's all getting too complex now."
Whether he's right, and this is the start of something new, or merely another new technique to add into the WAN acceleration mix, only time - and the experiences of users - will tell.