Riverbed's announcement this week that it is at last shipping its PC client software has underlined something that Techworld readers have known for a while - that PCs are now powerful enough to do the job of a WAN optimisation appliance.
That changes things - and it's going to change them even more.
Most importantly, it changes the value equation. In the past, WAN optimisation needed an appliance at each end because the storage and computation work involved was beyond the capability of an individual PC. Because it needed an appliance, it was only cost-effective when done for a number of PCs - a branch office, say - and because most of the applications in use were internal ones, that wasn't a problem.
Now though, any PC can do TCP optimisation, data compression, and some degree of CIFS caching, while Web 2.0 and SOA are breaking down the borders of application deployment.
The question then is what happens if a basic WAN optimisation client were included in the next version of Windows, for instance?
"I think Microsoft has already made moves in this direction," says Chris King, Blue Coat's director of strategic marketing. "It has improved the TCP stack in Vista with more flexible window sizing for example, and it has improved CIFS. It also hid the pain of MAPI so the user didn't see the slowness of downloading email.
"Now that Microsoft is improving TCP and CIFS, in a year or two there won't be any budget dollars in optimising those - although with TCP there will always be network issues outside the OS that you can work with, such as satellite latency.
"Plus, there will still be improvements you could do elsewhere, and a whole bunch of other apps will remain bulky on the wire, such as video and WebDAV, so over time we will shift onto other areas."
Could a New TCP be the answer?
One possibility would be for the WAN optimisation specialists to co-operate on extending TCP for the WAN, as it would enable them to move on from the basics that everyone needs to focus on the areas where they can really compete.
"The TCP latency mitigation piece definitely needs to be fixed," agrees Craig Stouffer, marketing VP at Silver Peak Systems. He adds though that this process is already underway: "TCP has been getting an incremental make-over, with multiple enhancements to the spec from various companies."
And he says that while the WAN optimisation vendors have mostly standardised on NetFlow for network reporting, it is unlikely that they would also standardise on an extended TCP stack. What's really needed is an industry initiative to get interoperable WAN-TCP specs in place, as a grounding for everything else, he says.
"There's TCP latency, and then there's compression, data reduction and all those other things - that's our real intellectual property," he explains.
Mark Lewis, Riverbed's European marketing director, agrees with the latter point, at least. He argues that TCP does what it was designed for, and that most of the optimisation today is actually in other areas, especially in improving how applications communicate over the WAN.
"TCP was designed for the Internet and it does a very good job," he says. "In fact the applications are the biggest contributors to network latency, and that's the responsibility of the application developers."
He adds that with little sign that they will properly re-write their apps for WAN use - "although there's always some optimisation going on" - there's still plenty of room for companies such as Riverbed to differentiate themselves according to how they optimise network traffic.
Lewis foresees this technology becoming "invisible" in the long run - Riverbed's software can already be deployed in visible or invisible mode, depending on whether or not you want the users to be able to see it's there and check what it's doing.
But he says that it will still be within a boundary, so a Riverbed client will only co-operate with a Riverbed head-end appliance.
Changing up a gear
Of course, even if the industry can agree on standards for some of that base technology, there will still be a need for WAN acceleration appliances in the datacentre because of the sheer volume of data going in and out.
Stouffer says that's why Silver Peak Systems has no plans for a PC client. Its main target is the backup and replication links between datacentres and the connections out to branch offices, not those to roaming laptops.
He suggests that the big issue for those offering PC clients will be deployment and support, and predicts that stand-alone WAN clients have little or no future, as their functionality will be taken over by VPN clients. Indeed, F5 Networks and Juniper are both believed to be adding WAN acceleration capabilities to their VPN software.
Even then, Blue Coat's King says that appliances will be more network-efficient than PC software once you have multiple users in a remote site. That's because if they work for the same organisation, it's likely they will be using the same files and applications over the WAN, so having an appliance means they can share its caching capabilities.
"Even for two users it's more efficient to have an appliance, though once you figure in the cost of the box, it becomes cost-efficient at around five users," he explains.
But he argues too that with more and more users going mobile, PC client software is no longer a nice-to-have - it is now a must-have.
"We're already winning deals and swapping out other suppliers because we have a client," he says. "I think that by the end of this year, if you claim to be a player in WAN optimisation then if you don't have a PC client you're done."