Crystal ball gazers trying to figure out what Cisco Systems Inc. (Nasdaq: CSCO) is planning for its future routers should take a look at a press release issued today by Corvil Ltd., an Irish startup (see Corvil Launches QOS Management).

Corvil is announcing some appliances that aim to take the guesswork out of capacity planning in IP networks. The bottom line is that they enable enterprises and service providers to cut costs by running their networks fuller than before. They also enable them to guarantee quality of service (QOS) when handling "next generation" applications such as voice over IP (VOIP).

This may ring some bells because it's what Caspian Networks Inc. claims for its flow-based routers -- a concept also espoused by another core router startup, Axiowave Networks Inc. (see Flow-Based Networking and Axiowave Queues in the Core).

Corvil's products aren't based on flow-based routing. They also aren't routers, but that's where the Cisco connection comes in.

Cisco is an investor in Corvil and speculation has it that Corvil's technology could end up inside Cisco equipment in the long run. Corvil CEO Donal Byrne gives oblique answers when asked about this, saying merely that the current appliances are the first way of deploying Corvil's technology in networks -- implying that there could be other ways in the future.

It's possible that Cisco might acquire Corvil in the way Cisco is acquiring Parc Technologies Ltd., a route-optimization startup in which it invested (see Cisco's Parc Purchase: An MPLS Play).

Corvil and Parc have complementary technologies. Parc improves the efficiency and performance of networks by showing planners how to spread traffic loads evenly over IP infrastructure. Corvil tells the planners where they're pushing the limits on balancing traffic, capacity, and performance. "They're all asking the same question: Where's the edge of the cliff?" says Byrne. In general (not specifically talking about Parc), Byrne says there's no point in using route optimization unless you can address this question. "It's like having a Ferrari with no wheels."

Cisco doesn't deny that Corvil and Parc are part of a grand plan. In the Light Reading article on Parc's acquisition, Cliff Meltzer, senior vice president for network management in Cisco's technology group, says: "There's actually more pieces to the story than just these two," and goes on to talk of a combination of in-house developments and outside investment that will deliver something "within the next 12 months."

Quite what that something is remains to be seen, but Corvil's inclusion appears to rule out Cisco going down the flow-based routing route. Corvil's technology has two key components:

First, Corvil's appliances sit at key points in a network and sample traffic volumes in a very granular way, every couple of milliseconds rather than every five minutes, as is the norm when routers collect traffic statistics. The millisecond measurement interval identifies the peaks and troughs of demand. Without this level of resolution, "it's like trying to be a microbiologist before they invented the microscope," says Byrne. The frequency of measurements has to be of the same order as latency requirements, which are in milliseconds for applications such as VOIP, he contends.

Second, Corvil has come up with mathematical algorithms that enable a large number of traffic measurements to be summarized in a single 160 byte message. These messages are sent to a central collection engine (a Linux server running Corvil software) every five minutes, where they're stored and served to Corvil applications running on desktop PCs. This avoids the network getting swamped with measurement traffic.

Caspian says its flow-based routing concept delivers equivalent "real-time utilization feedback" and that its method of looking inside every packet is equivalent to measuring traffic "in milliseconds or even microseconds," according to Dallas Kachan, Caspian's director of marketing.

Corvil's Byrne says high resolution measuring is one thing and being able to predict traffic conditions on networks based on those measurements is something else. No other technology can answer the following three questions that are uppermost in network planners' minds, he contends:

Corvil's press release makes a big thing of the "pioneering" mathematics behind its developments, but a researcher working on capacity planning at a major European incumbent carrier suggests this is overstating the case. The academic work cited by Corvil is "at least seven years old," writes the researcher in an email to Light Reading. "When it came out it did not have a huge impact on QoS research," adds the researcher, who requested anonymity.

Byrne responds to this by saying that the key research, concerning something called the "rate function" which characterizes queuing in mathematical terms, has not been published because it's part of Corvil's intellectual property.

Byrne also notes that Cisco was deeply sceptical of Corvil's claims to start with and spent three years investigating them before making its investment. "They started out saying there's no way this is going to work," says Byrne. "We jumped through all the hurdles."

Corvil is also able to demonstrate that its technology works in real life environments -- notably in enterprise networks, its initial target market. Light Reading has talked to the vice president of corporate technology strategy of one of Corvil's customers, a major financial institution in the North American mortgage market, who requested anonymity. His team is rolling out voice and video over the company's IP network and using about half a dozen Corvil appliances to help it forecast the bandwidth it needs to meet QOS goals.

Corvil's equipment has enabled the company to spot anomalies that hadn't been apparent when using other tools, says the VP. For instance, it showed that some types of traffic had been given an unnecessarily high priority. It also showed that in one case, a service provider had forgotten to throttle a T3 connection so the company was getting twice as much bandwidth as it was paying for. This helped explain why QOS was better than the Corvil equipment had forecast.

The combination of Corvil's equipment and bandwidth-on-demand from the service provider "is stimulating a lot of thought into really virtualizing the network layer," says the VP. It might help the company look at higher level issues, such as where it should host applications to optimize performance and minimize costs.

It's worth pointing out that Corvil has a sort of half-brother in the U.S., in a startup called CPlane Inc.. Both companies stem from a European research project into measuring and controlling traffic in ATM networks. The resultant intellectual property was split, with CPlane focusing on service control and Corvil focusing on measurement and QOS. It extended the work to cater for variable length packets.

While Corvil has won Cisco's support, CPlane is cited as a "partner" by Juniper Networks Inc. (Nasdaq: JNPR), Nortel Networks Ltd. (NYSE/Toronto: NT), and Redback Networks Inc. (Nasdaq: RBAK). A question mark hangs over CPlane's health. It hasn't posted any news on its Website for more than a year and phone calls to its offices went unanswered on Friday. One of its investors, Onset Ventures, says the company isn't dead and isn't hibernating, and has passed on a request to its CEO to tell Light Reading what (if anything) is going on.

— Peter Heywood, Founding Editor, Light Reading

For more on this topic, check out:

For further education, visit the archives of related Light Reading Webinars: