To improve broadband speeds, don't fiberise, optimise

Stuart Corner
Wednesday, 26 August 2009 06:39

IT Industry - Strategy

Dixon gave a real world example: "A large financial services company put in a one gigabit per second link between Sydney and Brisbane for data centre duplication but the maximum throughput they were getting was 24Mbps."

He added that the same company also had a problem with branch connectivity. "They were opening branches at a rapid rate but sometimes they could not get the leased lines into their buildings in less than three months. They found that plugging our [wide area network] optimisation appliance into a router with a 3G interface they were getting an [effective] 30-40Mbps. It gave them such high performance the did not need to consider putting servers in each branch." So over that 3G link they were getting better performance than over that one gig link.

Deployment and use of optimisation technology such as Riverbed's in networks that are open to public access would represent a radical departure from the way the equipment is presently used, but according to Dixon this would present no barrier to his company's technology because Riverbed appliances and software automatically discover each other if a communications session is established between.

"Our idea is that we would stick racks and racks of this stuff into data centres and if you happen to be a Riverbed customer and you access one of these sites the two boxes will discover each other and start optimising," Dixon said. "And the smart thing will be when we see it turning up in the PoPs of service providers."

However while Riverbed's technology incorporates multiple techniques for optimisation, including repackaging data into larger 'blocks' to reduce the impact of latency, it achieves its biggest gains by simply not sending most of the data.

According to Riverbed's web site: "Steelhead appliances transparently intercept and analyse all of your WAN traffic. TCP traffic is segmented, indexed, and stored as 'segments' of data, and the 'references' representing that data are stored on disks within Steelhead appliances on both sides of your WAN. Once the data has been indexed, it is compared to data already on the disk. Segments of data that have been seen before are not transferred across the WAN again; instead a reference is sent in its place that can index arbitrarily large amounts of data, thereby massively reducing the amount of data that needs to be transmitted. One small reference can refer to megabytes of existing data that has been transferred over the WAN before. Data Streamlining is highly scalable, with peak data reduction ratios that can be as high as a 100:1, and sometimes even higher."

This technique is likely to me more effective in a corporate network - where the same data in modified forms flows back and forth across the network many times - than in the public Internet.

Dixon acknowledged that Riverbed's technology had to have previously 'seen' content before to deliver the full optimisation, but suggested that its installation at 'staging points' in service provider networks, close to end users, would greatly improve performance simply through its ability to mitigate the affects of latency over the long haul sections of communications links.