Stardust Global Ventures

Sheryl and I were talking with our pal Dean Elwood the other day about unified communications, what it really means, and who the winners and losers will be. During our conversation, I had an idea come to mind about what’s really going on in communications that I’d like to expand on a bit. Settle in for a bit and follow along. I’ve got some stories to tell and some things for you to consider.

In part, this is an expansion of some ideas I expressed recently in The Flawed Delusion of Telco 2.0 and then again in More on the Death of the Telco Paradigm. And while those posts didn’t generate the conversation I’d hoped, it’s not a topic I want to let go of, but perhaps I can find a better way to make my point.

Our conversation got around to the question of who’s going to win in the unified communcations space. Dean asked if I thought the ultimate winner was going to be someone like Google or Microsoft, a large dominant player. In short, we speculated whether a return to some pseudo-monoply environment is the destination unified communications is headed for, or whether standards, APIs and interoperability are the real means to an end.

NOTE: Any and all company names used herein are purely for example and name recognition. I am neither disparaging nor promoting any company in particular by use of their name or logo.

I said that I think Google, Microsoft, Cisco and the like are unequivocally the most at risk of complete failure and implosion in the unified communications evolutionary path, and I’ll explain why. I promised stories, so let’s begin.

Back in the heyday of the PSTN, there was a major milestone event. In the early 1960s, a militia group blew up four microwave towers in Utah. This event cut off communications in the western United States. The circuit switched PSTN was a part of our daily lives, and we realized it was vulnerable. At the time Bolt, Beranek and Newman (BBN) had been discussing the Intergalactic Computer Network concept published by J.C.R. Licklider. This dialogue and research work led to Licklider being appointed to a new role at ARPA. One of the primary goals was to study packet-switched networks vs. circuit-switched networks. We had a problem with single points of failure in the telecommunications infrastructure.

ARPANET stared operation in 1969 with four nodes. Since then, it’s grown into what we call the Internet today.

We can reinforce the single point of failure dilemma with another story:

On May 9, 1988, an electrical fault started a fire in the Hinsdale, IL central office operated by Illinois Bell. Early during the fire, telephone services failed. The fire department didn’t arrive on site for 45 minutes. Because of the dense, black smoke, firefighters had difficulty entering the building and locating the source of the fire. Emergency power was automatically provided by generators and batteries and could not be shut off easily. Neither standard dry chemical nor halon extinguishers were effective against the fire. Water had to be used, which exposed the firefighters to electrical shock danger. It took firefighters more than two hours to shut down power, enabling them to control and extinguish the fire. It was over six hours after the first fire crews had arrived on the scene that the fire was declared under control.

This fire was confined to an area roughly 30 feet by 40 feet on the ground floor. Cables were burned to various degrees and smoke residue covered most of the ground and parts of the first floor. The most severe damage away from the fire was caused, not by flames, but by corrosive gases in the smoke. These corrosives damaged the equipment that survived the fire. While the existing equipment was cleaned and used to provide interim service, it was deemed unreliable. This equipment all had to be replaced over time after the fire.

This was another debilitating example of single point of failure in the PSTN. Any technologist will tell you single points of failure are bad. As network technologies have matured and become part of our language, we also view single points of failure in businesses or industries as something we should eliminate.

And if we think of this in terms of unified communications and the Telco 2.0 mindset, our network evolves. It’s important to think not of the PSTN vs. the Internet. That’s an irrelevant comparison. Rather we need to think right now purely in terms of major single points of failure. And yes, I know I’m simplifying this, but bear with me. It will really all make sense.

What’s different? I’m going to say nothing. In terms of single points of failure we still have major single infracture components that potentially impact the network in huge ways. If we settle on these three examples as the major providers of unified communications, we’re putting too many eggs into one basket from a unified communications business perspective.

If we think about the business, or suite of services we call unified communications, we must avoid a service business that relies too heavily on this type of infrastructure. It’s as basic as the difference between circuit switched-networks and packet-switched networks. For those who might not have a background in switching technologies, here’s an excerpt from IP Telephony Demystified:

Circuit Switching
As the name implies, circuit switching is a technique whereby the switches establish a dedicated electrical (or optical) path between devices. The important point here is that a path or circuit is established for the duration of the call and is dedicated to the call. These resources in the network cannot be used for other calls or by any other user until the call is completed and the resources are released and available. In the telephone network, the circuits are set up before the call is connected, then released after the parties hang up the telephone.

Because networks cannot be designed to support every possible telephone call at the same time, the switches are designed as “blocking” switches. This means that when all available resources are in use, callers will experience queueing delay or blockage until resources become available.

The delay through the network once a connection is minimal. Most of the telephone network has been designed to provide about 55 milliseconds of delay over the circuits that are established.

The two most common circuit switched networks in use today are the public switched telephone network (PSTN) and the circuit switched public data network (CSPDN). In the United States, services referred to as “Switched 56” would be an example of the latter. With the advances of optical networking technology in the past few years, there is also considerable speculation that switching optical circuits may become a common practice.

Circuit switching can be implemented in several different ways. Space, time and frequency each might provide the dedicated resource used in providing the path. Space division multiplexing is the oldest form used and provides a spatially unique path through the central office switch. In their earliest implementations, these switches were manual switchboards used in telephone offices. Operators manually plugged in cords to provide the necessary circuit connections. Step-by-step and crossbar switches are also space division switching techniques. And newer switches use computer programs to control electronic components.

Time division switching dedicates a time slot to the parties involved in a call. The two parties are guaranteed a time slot, or all of the bandwidth for a guaranteed portion of the time. In this method, the equipment used at each end knows what time slot to use and both ends communicate during the assigned time slot. Since these time slots may occur several thousand times per second, the human ear cannot detect the underlying technology. A more complex variation of this approach is used in many central office switches and PBX systems today.

Cellular telephone networks and cable modem systems use another approach called frequency division switching. In this approach, communicating parties are assigned a dedicated slice of the passband within the total bandwidth available in the communications channel.

The key to each of these approaches is that something is dedicated to the call for the duration of the call. This dedicated resource cannot be used by any other user on the network until the call completes and the resources is returned to a pool of available resources.

Because circuit switching guarantees a dedicated path that cannot be shared, it requires a significant engineering effort to locate, reserve and connect the necessary resources through the network. This drive the cost of a connection up and causes some delay in the setup process. As a result, circuit switching is more economical for connections of a longer duration like a voice telephone call where parties may talk for three or four minutes. It works best when network utilization of the network is high, providing a usage level that keeps the resources busy, but not overloaded.

Packet (Store-and-Forward) Switching
There are many ways to provide switching without the use of dedicated facilities. One example of a store-and-forward switching network is the subway system in New York. Passengers can travel between any of the subway stations along the route. The topology of this network is referred to as a “hub and spoke “ topology. The subway has many switching points or nodes. To get from one location to another, users might have to transfer from one line to another at one of these nodes. At the hub nodes, passengers (the traffic) might have to wait in a buffer (be stored) until the next available train arrives so that they can move (be forwarded) on to their destination. Just as passengers encounter delays in waiting for a train to arrive, and sometimes queuing delays when the trains are fully loaded, a store-and-forward type network provides service that has very different characteristics from a circuit switched network, which would be more similar to a New York taxi; dedicated to a passenger for the duration of the trip

In a data network, even the links between the switches are shared on demand. Switches perform routing calculations to determine which link to send the data onto, then the data is placed in queue for that link. Resources are allocated on a first come, first served basis, and there are no guarantees that the next leg of the path will be available upon arrival. Delays in queuing can cause data to sit in buffers. In a data network, this means that the delay through the network can be sporadic and unpredictable.

Because of this unpredictability, large blocks of information aren’t well suited to this type of network. Large blocks in information have to be broken into smaller chunks in order to not degrade performance of the network. When we think of a four-minute telephone call, we are really thinking of a very large block of data.

In a store and forward network, each block of data has to carry some form of addressing information that the switches can use to determine where the final destination is. Without this, the information can never be delivered to the recipient.

Data applications are often described as being “bursty in nature,” meaning that there may be lapses or pauses between transmissions. Unlike a voice call, which is a real time interaction between two people, a data connection is often an interaction between two computers without a person directly involved. Since store-and-forward, or packet networks use statistical multiplexing, or first in, first out (FIFO) methods, this type of network is better suited to a bursty type of traffic, like data.

Packet switching is the most common form of store-and-forward switching in use today, with routers being a perfect example of a store-and-forward switch. Packet switching breaks blocks of information into a pre-defined size or size range. This process of packetization does create some overhead, as each packet must have addressing information. Error checking can be performed on a per-packet basis, and if errors occur, only the corrupted packet needs to be retransmitted. This gains some efficiency in the network, as long messages do not need to be repeated entirely if an error occurs.

Connectionless vs. Connection-oriented Networks

The circuit switched network is clearly a connection-oriented network. The connection is the call setup process that establishes the circuit.

Packet networks can be either connection-oriented or connectionless. In a connectionless network, no setup is required. Each packet carries sufficient addressing or routing information to allow it to be passed from node to node through the network. Since there are no guarantees, because there are no dedicated network resources, these networks are referred to as “best efforts” networks, and performance may be unpredictable. In cases of network congestion, these networks will often discard packets to alleviate congestion or crowded buffers.

Connectionless networks also do not guarantee that packets will be delivered in the order they were transmitted, Packet might take different paths through the network and arrive at different times. This means that the device at the recipient must have resources to store the packets until enough have arrived to reassemble the message for delivery.

The postal network is a good analogy of a packet network with no guarantees. You could receive this book in many packets, each containing a page. The envelopes would be the packets containing the message. At the receiver’s end, the pages would have to be stored until all were received, then the book could be assembled. If a packet were damaged along the way, only the damaged page would require retransmission, not the entire book.

As you can see, packet networks provide a good technology for delivering short or bursty messages that don’t require the overhead of call setup. Transaction processes, like ATM machines or credit card verification terminals generate short messages that are ideally suited to this type of network.

So at the end of the day, both have strengths and weaknesses. The same applies to connectionless and connection-oriented networks. Each serves a purpose, and each has value…in its place and in its time. One thing we’ve learned as the Internet grew is that routing protocols today are resilient and route around failures.

In 2003 there was a major power failure on the east coast. It took power out to several major cities in the US, including New York, Detroit and even parts of Canada. Certainly user traffic on the Internet decreased during the outage. Millions of homes and business were without electricity. But the Internet – the service network – simply routed traffic around the affected areas. Users outside the directly affected geography were not impacted by the outage.

I think we’re leading up to the same approach for sustainable unified communications. Go look at Internet maps and you find many pictures that look like this.

The Internet is an intricate mesh that simply routes around problems as they occur. Isn’t that what we really want of unified communications?

So which is more powerful?

• A network of some major players representing potentially large single points of failure?
• A mesh of small players all connected and intelligently routing around failures?

For me the answer is crystal clear. The power is in the mesh.

If the major players represent the equivalent of a Class 5 telco switch in the traditional PSTN, the innovators represent a single router in the Internet. There’s a direct correlation in this analogy when we think about the loss of a single node.

I want my nodes to be resilient, and I want my unified communications service to automatically route around or compensate for failures. That cannot be achieved by a monopolistic view. The market dominance of any single entity carries an unacceptable level of risk. Consider denial of service (DOS) or sercurity breach. Even a simple code revision on a major network can create immeasurable user impact. Anyone who’s done development work will admit that sometimes a project can scale so large that it can’t be effectively simulated in a lab environment. I’d posit that Google can’t, in practical terms, test every change they make to a degree of certainty about user impact. At some point, the law of diminishing returns takes over and changes get rolled out into production networks. They have to in order to keep the wheels of progress turning.

A monopolistic approach is still behind the Telco 2.o mindset, and that simply cannot sustain our communications growth. The philosophy behind communications services must become the same philosophy that has let Internet technology succeed.

The winners in the race to unified communications are inevitably the small innovators, and they are legion. I won’t hyperlink, but let me toss a few names your way – Voxeo, Abbynet, iotum, Truphone, Jaduka, SightSpeed, Vidtel, MaxROAM, Solegy, Sangoma, Digium, MOBIVOX, Jott, Twitter, Gist, Salesforce.com, Jajah, Phweet, Jive Software, Dialogic, Acme Packet, fring, NetQoS, Fonolo, SpinVox, QIK, Covergence, Cubic Telecom, Facebook, Audiocodes, Sipera Systems, TelEvolution and many more.

The losers? That’s easy. The 800 pound gorillas are the losers. They have to be. It’s natural selection. The path they’ve selected has ensured they’ll be the losers. You can name names yourself. But just because they’re the losers doesn’t mean they won’t be profitable, good investments, and even grow. Losers can be profitable too.

I often refer to technology as plumbing, and it bothers me that so many of my colleagues take that as a negative statement. Plumbing is a profitable, sound business to be in. And the infrastructure of our water network is vital to our civilization. The same applies to the communications plumbers out there. They support and maintain a vital part of our information plumbing.

Whether you’re a winner or a loser, a plumber or not, being profitable in a sustaining industry delivering service isn’t a bad place to be.

Technorati Tags: unified communications, packet switching, circuit switching, networks, sustanable, profitable, service, delivery, winners, losers