Originally published by Interop This Week, reposted with permission.

IP is under siege again. In February’s issue of BCR, A.D. Little’s Goldstein and Solomon suggest enterprises “slow the mad rush to IP”, and maintain that while IP has proven itself capable of running over everything, attempts at running IP under everything “add complexity and cost with little or no benefit." Columns like this conjure Dana Carvey’s impersonation of George Bush, Sr.: “Bad IP, bad! Lossy, lossy, lossy. No good for voice, uh-uh. Bad for videos, bad.” ATM and Frame Relay. Kinder. Gentler.

As Yogi Berra said, this is déjà vu, all over again.

Déjà vu

I spent time on Team IP-Skeptic in the 1980s, promoting OSI, the  paper contender. (During the OSInquisition, I recanted, was reformed, and am now a true IP believer.) First at Interops in the 1980s, then at successive Networld+Interop Conferences, and against OSI and all subsequent competitors, IP proved it could provide seamless interconnection of WANs and LANs of diverse characteristics: efficient, scalable, tolerant of layer 2 shortcomings and excesses, and able to support any application.

IP is not the last woman standing by accident, and not entirely a Cinderella story.  I co-authored a book nearly a decade ago, and in it, chronicled the success factors for IP, then. It’s remarkable how applicable these remain. In 1993, TCP/IP’s overwhelming positives were:

• It is widely available and understood;
• It works;
• It has carefully protected the installed base; changes are introduced only after lengthy evaluation of the impact of IP networks already in operation

These remain as true today as nearly 10 years ago. What we did not appreciate in 1993 was the contribution that open source had and continues to make in accelerating innovation in TCP/IP. We mentioned “all those university students at *NIX workstations” as a resource to be reckoned with, but we underestimated the adoption of the open source philosophy across platforms. The investment community appreciated this asset enough to inject billions of dollars into IP technology in the 1990s. These capital investments directed hundreds of wünderkinder from their “academic toy networks” to practical and profitable engineering, and are in large part responsible for accelerating IP’s ubiquity.

“IP over everything” is as close to incontrovertible as technology can get. IP’s run over everything from extremely low bandwidth amateur packet radio in the early 1990s, to 1000 Base “barbed wire” Ethernet at Networld+Interop in 2000; and today, over fiber approaching terabit rates.

The “simplicity of running WANs over Frame Relay and ATM”? Simple for whom? Service providers still don’t understand that IP is a core competency they must acquire to compete in the new millennium. Slow the mad rush? A better course for telcos would be to apply Tauzin/Dingell (HR 1542) lobbying dollars to hiring IP talent, and doing so before they purchase IP-based equipment. Think Ethernet everywhere, including the local loop, don’t expect to amortize switching over 30 years, and build out your network while everyone’s expecting you lose money!

If IP has a fault, it’s that it succeeded too quickly in becoming a global data infrastructure over everything! Before you whine about jitter, loss, and World Wide Waits, look up how many years elapsed before telephony and television could claim ubiquity. ISDN’s been around nearly as long as IP; care to compare global reach and numbers of subscribers?

The real itch everyone wants to scratch is whether IP under everything makes sense.

Déjà vu, Round Two

IP’s sprint to ubiquity compounded expectations that it would become the unifying communications fabric, that it would accomplish what TDM and ATM could not at a pace equaling or surpassing Moore’s Law. That VoIP quality isn’t already equal to telephone and television network counterparts is a disappointment for investors and an opportunity for devotees of IP’s evil stepsisters to wave fingers and say convergence won’t come through IP.

That IP QOS hasn’t materialized in 10 years should hardly be an indictment, given the track record of its WAN predecessors. But hey, how often do you get to tease supermodels over blemishes? FR and ATM have QOS, but let’s pause to compare apples to apples. FR/ATM traffic engineering is designed on a whiteboard, and metrics to maintain paths with predetermined service qualities are manually configured into each switch; the process iterates when bursty data patterns change the game – like, when a new application arrives – which is rare for voice and video, but oh so frequent for data. On occasion, operations staff can break away from this task to reconfigure SONET rings by the same process.

My colleague and VPN Day co-instructor Joel Snyder comments, “With all their superior engineering, telephone companies insist on building data networks the way they built the PSTN, and they price service so high subscribers invariably seek ways to optimize and bypass. IP is not inherently best-effort delivery, but when you build it best-effort, it’s damned cheap technology.” Today, everyone engineers IP networks to be cheap, not fast nor reliable. Trying to run QOS-sensitive applications today over IP networks is tough because no one’s willing to pony up the cash to build an IP network as reliable as the PSTN. Provider, heal thyself.

I for one want more from convergence than to sit in traffic with integrated services on my cell phone. I want broadband and convergence. FR and ATM can’t deliver convergence to the desktop, because they’ll never do broadband data as well as IP and Ethernet in a “bandwidth consumption only rises” universe. IP can. For IP to support convergence, service providers have to either invest tons of money to over-engineer the IP network as they do the PSTN, or they must rely innovation beyond the reach of FR and ATM.

There’s no glorious ultimate contender on the horizon: IP’s the answer. But IP convergence isn’t here yet. So criticize the IP community for pursuing what cell-heads consider a Quixotic task of incorporating distributed, adaptive response into label switching and differentiated services protocols. Criticize them, too, for thinking about how this technique will work today, over any communications substrate, including optical, and over yet to be conceived substrates of the future. IP’s first three decades of innovation are enough to convince me that it’s a “can do” technology. It’s only a matter of time and talent.

And working capital.