Widespread access to networks, via WANs and the Internet, would seem to go hand in hand with the growing popularity of Web applications but that is not necessarily the case. Instead, many companies are finding that when applications are designed for use on a local area network long distance connections can pose serious challenges, whether they are wired or wireless.

Once the initial problem of sufficient bandwidth is overcome the bottleneck shifts to network latency. The nature of Web-based transactions means there is not much that can be done without redesigning the application. So says Nat Kausik, the CEO of Fineground Networks, a company whose software wraps around Web apps to make them WAN-friendly.

"The problem is latency, so compression or bandwidth management can't help. The typical symptom is HQ staff are happy but branch staff are unhappy, because of poor performance, for example five second response times versus 50 or 60 seconds. We could get that latter down to 20 seconds," he says.

"Caching helps with consumer-based applications but in the enterprise it makes little difference because applications tend to be SSL-encrypted or highly dynamic. The other option would be to bring in consultants to study the application, and the network, but spending money there won't cut the latency on the link to Hong Kong."

The problem is that Web apps and protocols are very chatty, he argues, with many transactions going to and fro - each one requiring a response. On a LAN this is not a huge problem, but on a WAN the inevitable delays are much magnified.

An increasingly popular example is the portal. Kausik says that one of his customers is aluminium producer Alcoa, which was considering installing a second Plumtree portal for its South American operations because its staff there were finding the main one unusable. They avoided that expense by using Fineground instead.

His argument is that the only practical and cost-effective route forward is to automate the way applications are optimised and this is what Fineground does. He adds that it runs in tandem with bandwidth-focused application optimation, with each technique addressing a different part of the puzzle.

"What we do is closely related to what Peribit and Expand do, they have a box in the datacentre and at each site to reduce the cost of bandwidth - that works for mid tier customers. But for the Global 1000 a box at each site is expensive and latency is still a problem. Ours is exclusively at the datacentre, it inserts itself and dynamically rewrites the packets going out.

"Our software wraps around the application. The application developer could do the same thing - people developing client-server apps used to write for performance - but it's a waste of their resources. Plus, enterprise application companies don't want to think about optimising for remote users because they don't want to limit their functionality."

It is not a cheap approach: Kausik says the typical sale is $80,000 to $100,000 and covers two Web applications, with each needing a separate licence for the Fineground software. He adds that the task includes monitoring software to classify the application and tell it how to tune it.

"It takes 20 minutes to install, then typically two days to configure it as part of a three to five day service engagement," he says.

"The design goal is to pare each transaction down to the minimum. One example is changed data elements - we encode just the differences in DHTML for any standard browser. We also minimise round trips by creating a single super-page."

He adds that a Fineground box can also accelerate single sign-on (SSO) schemes, packaging requests as one packet to avoid the need to continually re-send the user's SSO credentials. The software can also back off and reduce its acceleration if it detects that the back-end server is overloaded.

"The company started three years ago, as the market was spinning down. Enterprises were deploying Web applications by the dozen and expecting their employees to use them for everything," he says. "We started the company with one concept, but soon learnt there's no single solution for all problems, so we keep adding technology to do with various aspects.

"A lot of our technology is adapted from the microprocessor world, for example from the optimisation engine in front of a chip."

And he notes that Fineground is not alone in this area: "We are seeing a category of application development products appear - layer 7 decisions have to be made at the packet level. There are other start-ups such as Netscaler, also larger companies such as BMC and Tivoli.

"Ours is the most expensive product, but it's still pretty cheap compared to how much you would have to pay Accenture to re-engineer your software."

Is Nat Kausik right about Web latency? Are you using Web apps over a WAN and, if so, what's your experience? Air your thoughts on our forum.