Communications

Layers violations still useful for wireless VoIP

A few days ago I wrote about the observations, by Christensen et al., on vertical integration – when it's valuable and when it becomes a disadvantage.  Their point:  when your task presses the limits of the available technology, optimization is essential and vertical integration is best, as the parts that must work together, i.e. be jointly optimized, are better controlled under the same roof.

As technology improves, you can afford the overhead and obtain the benefits of modularity.  People working on separate modules don't need to coordinate details within their modules, separate companies can specialize, and really big systems (like the Internet) can emerge.

VoIP, i.e. telephony which is separate from the underlying transport, emerged once there was excess capacity in the underlying transport.  VoIP's first major success was in international wholesale (iBasis and ITXC – later bought by Teleglobe) where the VoIP carriers connected directly to the Internet backbone thus getting adequate capacity.  The second major success was IP-PBXs, running on enterprise networks that were over-provisioned and/or had simple QoS (typically 2 levels; gold bits & brown bits).  The most recent success is the emerging consumer VoIP  – Vonage, AT&T CallVantage and the like, and Skype!  These services had to wait for deployment of residential broadband.

Compare this with mobile phone services, mobile Internet and other forms of wireless Internet.  How close are we?  Unfortunately we have a way to go.

Yes, I've used Skype over wireless and, yes, the much hyped IP Multimedia Subsystem (IMS) is based on IP and SIP.  But it's not so simple.

WiFi networks have plenty of capacity if you are sitting in one place with only a few others sharing the same access point.  The moment you roam to another access point you face a gap in connectivity of perhaps several seconds, at least if it's a layer 3 handoff (to an access point with a different SSID) requiring a DHCP request (for a new IP address).  High traffic can cause even stationary web surfing to suffer.  Henning Schulzrinne recently presented work done with Andrea Forte and Sangho Shin at the 65th IETF meeting in March.  They measured hundreds of clients connecting with multiple 802.11 access points.  Among other things, as congestion went up the rate of handoffs soared – less than one minute between handoffs for 22.8% of the clients and between 1 & 5 minutes for another 34%.  The clients are continuously seeking stronger signals and end up thrashing between access points or even disconnecting and then reconnecting to the same access point.  The good news is that most of the problems they found can be fixed with better algorithms using existing 802.11 protocols and some clients, e.g. Apple's, already perform quite well.  But still it will take years to see new algorithms that are also widely deployed, i.e. a new generation of equipment.

With 3G mobile networks, we have different issues.  Yes, IMS is "VoIP", but it's highly optimized and tightly controlled (by the mobile operator).  Thus, neither Skype nor Vonage will use IMS without first partnering (i.e. paying) the mobile operator.

What about straight Internet access?  The good news is 3G networks have been designed to support Internet access, at least asymmetric access for web surfing, and have been optimized for handoff of active IP sessions.  Implementing handoffs has required protocol optimizations (network engineers call these layer violations) and throughput has been optimized via centralized management of client power and uplink scheduling (violating the end-to-end principal?).  Luckily, these optimizations happen below layer 3 – there really is a modular boundary at the IP layer.

The immediate problems are too little capacity on the uplink, at least during busy periods, long latencies and walled garden policies (whether implemented with IMS or otherwise).  With first generation W-CDMA equipment, users share 384 kbps (peak) on the uplink.  That's fine at 2am, but less so when there are others on the air.  Ubiquitous deployment of HSUPA will improve latency and provide adequate uplink capacity, e.g. enough headroom so explicit QoS is no longer an issue for 3rd party VoIP.  I'm guessing HSUPA hype in 2007, urban deployments in 2008 and widespread availability in 2010 and beyond.

Walled gardens are a big issue today, but one that will fall to competition.  Once there is widespread, broadband wireless Internet connectivity from competing mobile operators, WiFi hotspots, WiMAX and other approaches, it will be impossible to sustain a walled garden.  Fixed-line Internet connectivity gives people access to myriad niche services that can't be offered under the walled garden business model.

Widespread use on 3rd-party VoIP on the mobile Internet may lag it's use on the fixed Internet by 5-10 years, but it's coming – IMS or not.

Technology & Competition in Voice Disintermediation

While Clayton Christensen is best known for his 1997 book, The Innovator's Dilemma, I've also enjoyed his subsequent observations on value chains, modularity and vertical integration.  A common basis for both his original thesis and many of his subsequent observations is what happens as technological progress outstrips customer demands.

Early in a product's life cycle, technology needs to be optimized to have any prayer of delivering acceptable performance.  At this stage a vertically integrated organization is ideal, as it can manage the entire value chain and deliver the needed optimizations.  This was case in telephony from it's 19th century origins to the 1960s.  But eventually technology performance gets ahead of what's needed.  To quote from the abstract of a November 2001 article by Christensen, Raynor and Verlinden, Skate to Where the Money Will Be:

... as the underlying technology improves to meet the needs of most customers, companies begin to compete on the basis of convenience, customization, price, and flexibility. At that point, vertical integration is no longer an advantage – in fact, it quickly becomes a disadvantage. Different links in the industry value chain become modular, and the chain subsequently fragments.

We saw this in telephony in the 1970s as modular subsystems became available – PBXs, microwave radios systems (the "M" in MCI) and mini-computers – each in turn benefiting from integrated circuit technology.  At every stage,

profitability goes to the companies that own the interdependent links in the value chain--the places where everyone's still vying to satisfy their customers with ever-better product functionality.

That's key.  Standardized modular components, serving multiple verticals, like memory chips, disk drives or computer servers, become commodities.  They can be profitable for companies that focus on process and operations, like Dell.  But the high margins are reserved for areas in the value chain where technology still has to be tweaked.

Of course, Christensen is writing about products in relatively free markets and just assumes there is competition.  In telephony, that's an unwarranted assumption!  Indeed, productivity gains in telephony lagged those of the IT/computer industry by decades.  But there have been a few breakthroughs.

With the breakup of the Bell System in 1983, we entered a period of long distance competition where the service caught up with what technology had already enabled – the death of distance.  Of course this also resulted in the demise of the long distance companies, as the capital investment to duplicate their networks was a small fraction of their sunk costs and the operating cost (of a new network) likewise a small fraction of theirs.

While there was no direct competition in local phone service until the recent advent of cable telephony and consumer VoIP, substitution of (competitive) mobile-only service has been a US trend since 2000.

Now VoIP is poised to do for voice telephony what long distance competition did for distance – kill it, at least as we've understood fixed voice telephony for the past 100+ years.  There will still be a PSTN interconnect business, but if both ends have broadband, the operating costs to connect VoIP phone to VoIP phone are near zero.  Today Skype has set the price point.  Monopoly (or duopoly) broadband providers may be able to sell bundles, but Vonage and other pure VoIP providers will be hard pressed to maintain a business in traditional telephony.

Christensen's model applies to many other areas, for example, layer violations in wireless protocols.  I will discuss the his model and the wireless industry in a subsequent post.

Can you trace Skype calls?

I won't rule it out, but what's being reported so far seems bogus.

Yesterday, Bruce Schneier pointed to this report on Ars Technica about Kobi Alexander, a founder of Comverse but recently a fugitive, purportedly being nabbed as the result of a one minute Skype call he made from Sri Lanka.  Since Comverse has been a major customer of NMS for many years, I've naturally been following their problems with stock option issue dates, Kobi Alexander's resignation and so forth.  I also follow Bruce Schneier's blog, as he has the best insights on security issues of anyone I know.  But this time, I think Bruce pointed to the Ars Technica article without actually reading the paper it cites.

The paper Ars Technica cites, "Tracking Anonymous Peer-to-Peer Calls on the Internet" by Wang, Chen & Jajodia of George Mason University, is very interesting and completely reasonable.  But what they are doing is correlating the timing of packet dispatches and arrivals at either end of a Skype connection to determine if two ends - both of which are already being monitored - are indeed talking to each other.  They conclude that, if they have access to 90 seconds of packet timing information from each end, they can reliably determine if two Skype clients are talking to one another.  I don't see how this could apply to Kobi Alexander placing a "one minute" Skype call from Sri Lanka.

If there is any truth to the report that he was traced due to a phone call, it's more likely the call was Skype to PSTN via SkypeOut, the PSTN leg was tapped (as discussed by Phil Wolfe) and the listeners heard him discussing his current location.

The paper by Wang et al has a very nice set of references (up to date through 2005) to other work on the subject.  On other fronts, there was a flurry in the blogosphere in July when a Chinese group announced they had reverse engineered the Skype protocols.  Reverse engineering object code is certainly feasible, although we've heard nothing further from the Chinese group since the original announcement six weeks ago.

Skype is becoming significant (eBay claims Skype handles 7% of all international minutes) and it uses proprietary protocols, not subject to widespread inspection by the security community, so it's reasonable to suspect there are flaws and that Skype calls will become (or already are) traceable.  But I don't think we'd heard a viable explanation of how to do it, yet.

Taxes, bureaucracy and corruption

Catching up on somethings I printed (in June no less), I read this study by the Telecom Regulatory Authority of India comparing the finances of the telecom industries in China and India. 

There a lot of interesting data, but some of their conclusions are amazing.  For example, in comparing mobile operators' OPEX per subscriber, listed as China $4.73 and India $5.49, they offer just one potential explanation

Economies of scale ... could be one of the reasons for lower OPEX for Chinese companies.

Funny, my first thoughts were:

• India's substantially higher taxes on telephony
• India's extra bureaucracy - In ease of doing business China ranks 91 and India ranks 116.
• India's extra corruption, ranking 88 to China's 78.

China might have economies of scale, but that's the least likely explanation, as China Mobile and China Unicom operate with highly regional decision making.

Optimism

This comment by Brian Thomas in Tom Evslin's Fractals of Change is a nice summary of my view of the Internet:

My personal belief here is that given equal exposure, the good generally wins over the bad, and better communication makes it easier for the exposure to be equal, so even though the Internet enables many new ways to degrade, defraud, and deceive people, it also provides new ways to expose evils.

There are many historical examples of this, from the invention of the printing press, then movable type, then telegraphy, telephony, radio, and finally the Internet. Gutenberg did more to break the power of medieval craft guilds and religious authorities, both of whom wanted to keep their knowledge secret.

Power law vs log-normal with really big bucks implications

Yesterday's post was rather rushed as I was about to get on an airplane.  Once on board, it occurred to me, there's another book I've read where the difference between log-normal and power law distributions is key.  That's The Misbehavior of Markets by Benoit Mandelbrot & Richard Hudson.  They argue modeling markets with normal distributions (in price changes, for example) is erroneous and leads economists to underestimate the likelihood of extreme events (like Black Monday in October 1987).

Put another way, small price changes occur very frequently and large price changes very infrequently.  If you plot these on a log-log scale you get a straight line out some distance. Conventional finance assumes this fits a log-normal curve which eventually falls off.  Mandelbrot, argues historical price data fits a fractal or scale-free model, i.e. follows a power law, meaning the likelihood of extreme events has been consistently underestimated.

Those of you in high tech in the US may have followed the accounting change which requires stock options be expensed by the company issuing them.  To do this one must be able to estimate the present value of the options are the time they are granted.  This is done using the Black–Scholes formula.  While the authors received the 1997 Nobel prize in economics (Scholes & Merton; Black had died in 1995) and the Black-Scholes formula is now written into accounting practice, Mandelbrot argues it consistently underestimates infrequent cases of extreme price volatility. 

As an interesting aside, Long Term Capital Management (LTCM), a hedge fund with Scholes & Merton on it's board, collapsed in 1998 due to a series of correlated repricing events that were extremely unlikely, but did occur in the late summer 1998.  LTCM lost $4.6B.

Power laws and log normal distributions appear in many disciplines, not just networking and markets.  In many cases it's hard to determine which applies.  Whether Mandelbrot is correct about power laws and finance, at least we can agree -- $4.6B is big bucks!

Scale free networks. long tails and big bucks - or maybe not

Some of my blog posts serve to organize a set of references on a topic that's important to me.  This is such an occasion -- hopefully one or two others may share this interest.

Back in 2002, I really enjoyed Albert-Laszlo Barabasi's book on networks, Linked: How Everything Is Connected to Everything Else and What It Means, but haven't paid much attention to scale-free networks since then.  Yesterday I stumbled on this interesting presentation by Chris Anderson at "SciFoo" last week, which prompted me to revisit the subject.  The notes in Anderson's presentation include two great references (Limpert et al., Mitzenmacher) for the mathematically inclined, and the current Wikipedia article on Scale-free networks has excellent content and a lot more references.

New to me, are the diverse models that generate networks which are scale-free but differ in other meaningful characteristics.  Barabasi discussed preferential attachment, a model which generates scale-free networks with highly connected nodes near the core.  These networks are robust even if a large fraction of the nodes are removed.  However the Internet, while a scale-free network, has it's highly connected nodes closer to the periphery, so the preferential attachment model doesn't apply.  And as Wikipedia currently puts it,

Indeed, many of the results about scale-free networks have been claimed to apply to the Internet, but are disputed by Internet researchers and engineers.

The "big bucks" part of my headline was prompted by Chris Anderson's The Long Tail: Why the Future of Business Is Selling Less of More.  Anderson's thesis, originally published in Wired magazine, is that the popularity of items in certain markets, like media, follows a power law -- the same law that applies to the connectivity of nodes in a scale-free network!  In such a market, a few products have enormous sales but there is a very long tail of products that have only a few sales each.  This long tail goes way out, so taken together, sales on the long tail represent big bucks.

Since Anderson's book was published there's been considerable argument about the long tail.  Nicholas Carr has good summary of the argument between Wall Street Journal columnist Lee Gomer and Chris Anderson.  But Andrew Odlyzko (on a private economist list, quoted here) has the best overall answer to the questions Lee Gomer raised.

As with the Internet, the key question is: does the phenomenon follow a power law (as scale free networks do)?  If however, it follows a log-normal distribution, there's a lot less money to be made selling products out on the long tail as illustrated by this slide in Anderson's presentation:

As Anderson puts it:

But here’s the problem. I’ve been assuming that the powerlaw is the “natural” shape of all these markets. But there are other distributions that start off straight on a log-log chart, then fall off the line not necessarily because of a bottleneck effect but because they’re simply not powerlaws. The best example of this is the “lognormal” distribution, which often stays straight for several orders of magnitude before sloping downward. How to tell whether a market that falls off the line is a natural powerlaw shape distorted by a removable bottleneck or a natural lognormal market that will look like that no matter what you do? In markets such as film or television archives, this could be a billion-dollar question.

Indeed!

Social networking just the platform for communities

As I've written before, social networking sites are worthless if they don't succeed in building communities, so I was struck by this report by Christine Herron on the story of Flickr, as told by co-founder Caterina Fake at the recent BlogHer conference:

The Flickr team spent a lot of time greeting every user that came to the site, offering them help on how to get engaged with the rest of the community. Through this, they discovered that online community-building is just like being the host of the party - if guests come to a party, and they don't know anyone, and no one shows up to take their coats and introduce them around, they'll leave. These social networking skills are essential practices to bake into community management. Eventually, you'll want people to identify enough with the community that they themselves will act as the community police, kicking out trolls and making new users feel welcome. This community will also provide much of the user education in a socially networked site - people tend to come once invited there by a friend for a specific purpose, so they come with some idea of what they will be able to do and how the site works.

Mixi.jp social networking going public

While I'm on the subject of social networking, Kenji Mori reports that mixi.jp, a major Japanese social networking site (Wikipedia info here) is going public on the Tokyo Stock Exchange in September.  While the potential valuation (over $1B US) seems absurd, at least as a public company we should get some insights into the inner workings of one social networking company - all the others I'm aware of are private of captive of a larger organization.

Fukumimi is already dissecting their prospectus (and reporting in English!).

Cyworld US goes live, but who will use it?

The major, major, Korean social networking site, Cyworld, has versions in Chinese and Japanese and, beginning today, in English.  The US site has been in alpha for many weeks, but today it's open to anyone, so I tried it out.  My new "Minihome" is here.

Looking around, i.e. sampling other members via the "Random Minihomes" button, the current US members are predominately Asian Americans.  Quite a few members list their MySpace and other contact info prominently on their Cyworld Minihome, but this may be a reflection of the alpha test policies and the familiarity of the Cyworld brand in Korea and Chinese communities.

A key feature of Cyworld in Korea, mobile access, is either missing or downplayed on the US site.  That's too bad.  Other than that, it was easy to get started and I think I prefer the look of my Minihome to that of my entry on Myspace.

If Myspace didn't already exist, Cyworld US might have a chance, but to go against Myspace at this point they will need a distinctive product position.  I don't see at it, at least at the moment.  As I commented in July, Facebook is taking the elitist position and thus growing their niche of college kids and college-bound high school kids.  Cyworld has an enormous, profitable business in Korea and they are owned by SK Telecom, so they have the resources to stick it out.  I'll be interested to see how their product position evolves.

Telecom growth in emerging markets - mid-year check-in

Pakistan continues to lead the world in mobile phone growth rate and China continues to lead in absolute number of subscribers added per month, but Indian's mobile sector is going great guns putting India close to China in absolute adds and #3 for percentage growth.

This update from Mobile Pundit caused me to do a quick check-in on some of my favorite emerging telecom markets.  Computing growth by comparing the most recent figures I could get (June or July 2006) with subscriber numbers from 3Q05, here are the current figures:

	Millions of Mobile Subs	Mobile Teledensity	Growth Rate
Pakistan	34.5	20.8	278%
Bangladesh	11.6	7.9	211%
India	108.0	9.9	202%
Nigeria	24.0	18.2	192%
China	426.4	32.5	158%

I've included Nigeria as it's the fastest growing telecom market in Africa.  As a continent Africa lags Asia, but it's a fascinating place, both for problems and opportunities. And of course, our AccessGate RAN optimizer is being widely adopted in Africa, so I'm paying more attention.

While Pakistan, Bangladesh, India and Nigeria are all stuck with relatively small and slowly growing fixed line networks, China is different with a large "fixed line" network undergoing significant growth.

	Millions of Mobile Subs	Mobile Teledensity	Growth Rate
China -fixed	365.3	27.8	78%

Some are these are actually fixed, some are WLL, e.g. the PAS or Little Smart system.  In any event that gives China a total of 792 million telecom subscribers and a total teledensity of 60.3, way ahead of other emerging markets.

Here's the rest of the data if you're interested.  Download mobile_teledensity_growth_as_of_mid2006.xls

I continue to be dismayed by bureaucracy in India, censorship in China and Pakistan, and corruption everywhere, but especially in Bangladesh, Nigeria and Pakistan.  But these numbers give me hope.  Telecom directly drives economic growth and diverse communication channels are the best antidote for closed societies.  It may take decades, but substantial parts of the world are headed in the right direction.