IP reliability to the rescue

About InfoWorld : Advertise : Subscribe : Contact Us : Awards : Events : Store

NEWS

IP reliability to the rescue

By Scott Tyler Shafer
May 14, 2002 1:01 am PT

See correction below

ADVERTISEMENT

<a href="http://ad.doubleclick.net/jump/idg.us.info.news/article;abr=!ie;pos=imu;pkey=business;skey=;paud=;saud=;sz=336x280;tile=2;ord=212304240708?"> <img src="http://ad.doubleclick.net/ad/idg.us.info.news/article;abr=!ie;pos=imu;pkey=business;skey=;paud=;saud=;sz=336x280;tile=2;ord=212304240708?" border="0"> </a>

Free IT resource

Hear how top CIOs turn change into a competitive advantage.

TOP NEWS

IT SOLUTION SEARCH

THE NETWORK equipment vendor's latest pitch to secure additional customers involves making comparatively inexpensive and unstable IP networks as reliable as their circuit-based voice counterparts.

Renowned for their resiliency, voice networks are still used by most carriers to offer much of today's voice services. But given the impending move to run voice and data traffic simultaneously over a single network, along with the carrier's mandate to cut costs and increase revenues, the IP reliability issue is right on time.

Cisco Systems and Alcatel are among those making the most noise about IP reliability. The two separately have announced recent improvements to their equipment's OSes that will enable carriers to ensure almost zero packet loss over their networks.

The technologies are said to achieve this by tuning the routing protocols that IP routers use to communicate with one another. By doing so, they will no longer drop packets and will have improved capability to switch over traffic when a failure occurs, without an interruption in service. These advances are enticing to most carriers, but each is on a different schedule in adopting them.

"Voice networks and their equipment are the most reliable thing on the planet," said Dave Garbin, vice president of network strategy at Cable & Wireless (C&W), based in London. "They never stop working and there has been only one failure in 40 years. IP has to live up to those standards."

Garbin explains that in the world of data services, ATM and frame relay technologies attempted to follow the reliability paradigm but were extremely complex and expensive. And Internet routers were not designed with the same reliability as voice equipment and thus rely on lower-quality recovery mechanisms to repair themselves, which they do well but not quickly.

Garbin estimates that a down IP network typically takes several minutes to several hours to come back up. On the other hand, voice networks recover in milliseconds.

But insufficient recovery time may no longer be an issue for IP networks. Cisco claims its new GRIP (Globally Resilient IP) technology can provide fast fail-over in MPLS (Multiprotocol Label Switching) within 50 milliseconds -- on par with voice.

This news is of interest to C&W, which is looking to migrate its network to MPLS, which creates paths over various link-level technologies such as Packet-over-Sonet, frame relay, and ATM, as the company prepares for converging voice and data.

Garbin explained that if a failure occurs today in the network, the core routers, which run over an ATM network, would have to find their own backup paths. This would result in minutes of down time and packet loss. A Cisco-enabled MPLS solution creates logical paths, allowing the carrier to specify a primary and backup in case of a failure.

Late last month C&W announced a four-year deal with Cisco in which the company will build out a network that takes advantages of GRIP. C&W will offer IP-based VPN and LAN services with guaranteed levels of QoS (quality of service) that rival that of voice.

At WorldCom the story is a bit different. To run UUnet, the world's largest IP network, WorldCom takes a more conservative approach to new technologies, especially those that promise improved reliability.

According to a WorldCom representative, switching customers to IP from frame relay or ATM-based services for the sake of a new technology is not of interest to the conservative carrier. She explained that WorldCom would rather not take the risk and moreover, its customers aren't asking for IP because many aren't looking at underlying technologies, such as switching, they are looking for high service levels.

However, WorldCom does have interest in next-generation routing. The company claims to have routers from most vendors in its labs and is interested in technologies that are scalable, reliable and interoperable. WorldCom buys from multiple vendors and requires that standards are being followed by its routing suppliers.

Meanwhile Cogent Communications, a provider of metropolitan bandwidth to enterprises and service providers alike, says reliability and QoS are of utmost importance. The Washington-based company sells bandwidth of 100Mbps for a flat monthly rate of $1,000.

On the IP-reliability front, Cogent contends that a properly engineered IP network can already outperform a circuit-based network. Dave Schaeffer, CEO of Cogent, said that many providers have built their IP networks over voice networks, which adds a level of complexity that may result in increased reliability.

"If you build an IP network from the beginning, you'll get better reliability results," said Schaeffer, whose company runs a pure Layer 3 network over Cisco equipment. "Reliability comes from path diversity."

Besides being unproven on a large scale, IP-reliability technologies still only work in a single-vendor environment. Both Cisco and Alcatel's technologies only work only over their respective owner's technology.

IP reliability is a step forward toward becoming the basis for converged networks of the future. But for now, carriers will be slow to carry more robust, mission-critical applications such as voice and call centers over IP.

"Whether a carrier can replace ATM and put services over MPLS-based IP networks is an argument down the road," C&W's Garbin said. "It will happen only if carriers can carry over IP and still be credible."

Correction

In this article, we misquoted Cogent's Dave Schaeffer regarding the result of providers building IP networks over voice networks. The practice adds a level of complexity that may result in decreased reliability.

	Who cares about IP reliability? Fueled by a big-bucks public relations campaign, IP reliability, some vendors say, is the next big thing. The Internet Protocol, these vendors say, isn't reliable enough for banks, the government, and other big players; something new is needed. Meanwhile, the rest of the world, after using IP for years, is wondering why their familiar network protocol suddenly isn't good enough. First, to dispense with the hype, the talk about IP reliability is aimed almost exclusively at carrier-class customers with converged networks. These are customers that today are using ATM or frame relay networks. Most other businesses using IP networks for moving data (not voice) traffic aren't really the targets of this technology. Nor is this technology aimed at companies that use the Internet as their means of connecting to the outside world. The technologies needed to put the required non-stop routing into practice are proprietary. Unless your network consists of hardware from only one vendor, you probably won't be able to use the promised products anyway. These technologies require special routers that you're not going to see widely deployed. In addition, much of what Cisco, Alcatel, and others are promising, such as the ability to route around an outage with no packet loss in less than a tenth of a second, is technology that already exists from other vendors. If you install such technologies, you would need to create a meshed network using only routing or switching nodes from that single vendor. This is not unlikely with carriers, but it's unlikely in the enterprise. Once you get past the proprietary nature of such networks, what you have is a connection-oriented version of an IP network. Normally, IP networks are connectionless, meaning you can send out IP packets without knowing whether they actually reach their destination. When they don't get to the other end, you can lose packets, and that's what recent IP reliability efforts are designed to prevent. But packet loss doesn't matter very much. When a stream of IP packets arrives at their destination, the packets are assembled into their original order, and missing ones are requested and inserted. This works fine most of the time, but not for voice traffic. For most voice traffic to be useful, it has to be delivered in a contiguous stream with little latency. To delay the data stream to request missing packets would cause too much latency, so the data stream containing the voice information must be sent along as it arrived. This is why voice over IP sometimes has a burbling quality. So for carriers planning to use IP networks to transport voice traffic, reliable networks such as proposed by Cisco and Alcatel make sense, despite their expense and proprietary nature. In other words, unless you're a carrier, or you depend heavily on converged voice and data networking, the IP-reliability hype has little to do with you. This technology is designed to make the phone companies feel better about IP; it's not about mainstream networking. But since no products have actually shipped, it's impossible to know exactly what the vendors will deliver. -- Wayne Rash