“Client-aware cloud” is one of the buzzwords that Intel was pushing at IDF 2011, and despite the chipmaker’s ulterior motive for wanting to see synergy (read: “vendor lock-in”) between (x86-powered) cloud clients and (x86-powered) cloud servers, there’s definitely something to larger idea that the cloud should know something about and tailor data for the clients that are connected to it. HTTP proxies like Amazon’s Silk and Opera Mini before it (not that these are the same thing) are part of this client-aware cloud trend, as is Skyfire‘s new Rocket Optimizer 2.0 service.

Skyfire is probably best known as a maker of mobile browsers, and specifically as the company that snuck porn Flash onto the iPad via some clever transcoding and a proprietary browser. But in an echo of Google’s 411 strategy, in which the search giant launched a free directory service in order to gather the voice samples that would later power Android’s voice recognition features, Skyfire has been using the data that it gathers from its 10 million iOS and Android users for a larger R&D project.

That project is Rocket Optimizer, the 2.0 version of which launched on Tuesday. Optimizer is aimed not at consumers but at wireless carriers, where it takes tower- and mobile device-specific data and uses it to transcode and optimize video data on-the-fly to fit users’ screens and network conditions. As with Silk, all of Rocket’s real-time video transcoding and resizing is done in the cloud.

Rocket is currently deployed with a tier 1 wireless provider, which uses the platform to look across its wireless network and identify cell towers that are congested due to a large number of simultaneous video downloads. Rocket is able to detect that congestion and optimize the video going through that tower and tailor it for available bandwidth and for each user’s particular screen resolution.

“The net result is that there’s a lot of pollution in these networks, and we can take out 60 percent of the bandwidth on average,” Skyfire CEO Jeff Glueck told Wired.

Unlike Silk and Opera Mini, no part of Rocket Optimizer lives on the client, and the platform processes only a small fraction of overall web traffic, ignoring the rest. This latter fact is key to Skyfire’s ability to do the otherwise impossible task of real-time video transcoding for massive amounts of video traffic.

In a widely cited study, Cisco claims that 52 percent of the Internet’s traffic is now video, but Skyfire’s user data indicates that only 2 percent of HTTP GET requests initiate the viewing sessions that account for all of that video bandwidth. So Skyfire focuses on filtering out everything but that smaller slice of HTTP traffic, which means that the company’s optimization platform doesn’t sit in between the user and the server for the other 98 percent of HTTP requests. (Contrast Silk and Opera Mini, which are true proxies that intercept 100 percent of HTTP traffic.)

With this targeted approach, Optimizer is able to call on the cloud portion of its platform first to monitor the load across a carrier’s network in search of congestion, and then to transcode video in real-time in order to cut down on bandwidth usage. The massively parallel, networked transcoding engine can handle things like audio adjustment, framerate adjustment, compression, and resizing, all with a latency of under 200ms so that the user’s playback experience (e.g., seek, pause, rewind) and the publisher’s experience (e.g., analytics) are not affected by the intervention. When the network is at peak congestion, Optimizer does full-throttle optimization, and when the load is low it does light optimization.

So by paying closer attention to which clients are doing what and where, Skyfire’s client-aware approach lets them do the seemingly impossible—real-time Internet video transcoding and compression—with minimal impact on the client or the network.