We believe that large clouds could one day deliver computing power at up to 80% lower cost than small clouds.  This is due to the combined effects of three factors:supply-sideeconomies of scale which allow large clouds to purchase and operate infrastructure cheaper; demand-sideeconomies of scale which allow large clouds to run that infrastructure more efficiently by pooling users; and multi-tenancy which allows users to share an application, splitting the cost of managing that application.

Ballmer appears now to get it. “The more searches, the more you learn,” he says. “Scale drives knowledge, which can turn around and drive innovation and relevance.”

Microsoft decided in 2008 to build 20 new data centers at a cost of $1 billion each. This was a dramatic commitment to the cloud. Conceived by Bill Gates’s successor, Ray Ozzie, the global platform would serve up a new generation of Web-based Office applications dubbed Azure. It would connect video gamers on its Xbox Live network. And it would host Microsoft’s Hotmail and search applications.

The new Bing search engine earned quick acclaim for relevant searches and better-than-Google pre-packaged details about popular health, transportation, location and news items. But with just 8.4% of the market, Microsoft’s $20 billion infrastructure commitment would be massively underutilized. Meanwhile, Yahoo, which still leads in news, sports and finance content, could not remotely afford to build a similar new search infrastructure to compete with Google and Microsoft. Thus, the combination. Yahoo and Microsoft can share Ballmer’s new global infrastructure.

I think that new video applications require lots more bandwidth — and, equally or even more important, that more bandwidth drives creative new applications. Two sides of the innovation coin. And I think investment friendly policies are necessary both to encourage deployment of new wireline and wireless broadband and also boost innovative new applications and services for consumers and businesses.

But this article, as one of many examples, mis-summarizes my view. It uses scary words like “apocalypse,” “catastrophe,” and, well, “scare mongering,” to describe my optimistic anticipation of an exaflood of Internet innovations coming our way. I don’t think that

the world will simply run out of bandwidth and we’ll all be weeping over our clogged tubes.

Not unless we block the expansion of new network capacity and capability.

The next line of attack is to use the words of the expert analyst Andrew Odlyzko to “debunk” a point they have just invented.

According to Odlyzko, the current Internet growth rate of about 50% per year “can be accommodated with essentially the current level of capital investment.” 

But this entirely misses the point. Why? Because the “current level of capital investment” is absolutely huge. And it is not a given — it is not automatic — that dozens of communications companies in a highly competitive, fast-moving broadband world can sustain the “current level of capital investment” if big obstacles are thrown in their path. In fact, it’s highly improbable.

In 2008, the U.S. invested $455 billion in information and communications technology. Things like computers, software, routers, switches, fiber optics, and cell towers. That was about 22% of all U.S. capital investment and an amazing 43% of all U.S. non-structure capital investment. We invested $127.3 billion in communications equipment alone (real private fixed investment; 2000 constant dollars). Communications service providers account for the bulk of this, investing about $70 billion per year.

It’s easy, with useful and entertaining Internet innovations popping up all around us, to just assume these modern miracles will continue flowing. But new productivity-enhancing technologies like cloud computing do not just appear out of, er, thin air. They require hundreds of billions of dollars worth of risky investment in support of speculative business plans in markets that are changing faster than any markets ever have.

The most reliable way to block the future is to assume it will happen. We have to make it happen.

But the content of the Analysys Mason (AM) study, entitled “Delivering High Quality Video Services Online,” does not support either (1) the media headline — “Much ado about nothing,” which implies next generation services and brisk traffic growth don’t require much in the way of new technology or new investment to accommodate them — or (2) its own “finding” that QoS and traffic management aren’t needed to deliver these next generation content and services.

For example, AM acknowledges in one of its five key findings in the Executive Summary: 

innovative business models might be limited by regulation: if the ability to develop and deploy novel approaches was limited by new regulation, this might limit the potential for growth in online video services.

In fact, the very first key finding says:

A delay in the migration to 21CN-based bitstream products may have a negative impact on service providers that use current bitstream products, as growth in consumption of video services could be held back due to the prohibitive costs of backhaul capacity to support them on the legacy core network. We believe that the timely migration to 21CN will be important in enabling significant take-up of online video services at prices that are reasonable for consumers.

So very large investments in new technologies and platforms are needed, and new regulations that discourage this investment could delay crucial innovations on the edge. Sounds like much ado about something, something very big. 

On the QoS question, AM appears to make a number of important oversights/omissions. I have only done a quick reading, but upon first examination: 

– AM considers only downstream non-interactive video, such as download-and-cache, streamed medium quality services like YouTube, and high-end video like IPTV. 

– AM does not consider two-way interactive real-time video, such as obvious applications like video-calling, video-conferencing, and Telepresence, or less obvious but still important future applications like 3D gaming, Cinema 2.0 interactive movies, high-resolution interactive virtual worlds, and a whole host of cloud-based services we cannot even imagine. 

– These real-time interactive video applications and services impose more severe restrictions on bandwidth, latency, and jitter than static downstream video applications. 

– The report does not appear to address the importance that QoS will play in the wireless/mobile realm. In the necessarily more bandwidth constrained environment of wireless, QoS is like to play a very important role in delivering video-calling, gaming, and other interactive services. 

– Throughout the report, AM also acknowledges the key role content delivery networks (CDNs) play in delivery of video and other content. CDNs are NOT a bandwidth solution. They are a caching solution that improves “QoS” through hop-reduction and latency-reduction. They are in fact a bandwidth-saving device, and also a topological mechanism to cope with the, alas, finite speed of light. 

– The mere fact AM spends so much time on CDNs undermines its own headline saying we can accommodate all new video traffic with more bandwidth. 

– Bandwidth is a very big part of the equation. But so are CDNs, QoS, and a range of other technological, topological, and architectural strategies. 

– I also have not yet found in the paper a detailed or persuasive rationale for the “finding” that bandwidth is ALWAYS a better solution than QoS. Bandwidth is indeed a good substitute for switching and/or QoS, but not always. And it is not always the most economically efficient solution. 

I may revise and extend these observations, but upon first glance, the headlines don’t reflect the substance, and in a few cases, like its analysis of QoS, substance doesn’t support the findings.

Almost 40 years ago the CPU — or central processing unit — burst onto the scene and enabled the PC revolution, which was mostly about word processing (text) and simple spreadsheets (number crunching). But today, as Nvidia and AMD’s ATI division add programmability to their graphics chips, the GPU becomes the next generation general purpose processor. (Huang briefly describes the CUDA programmability architecture, which he compares to the x86 architecture of the CPU age.) With its massive parallelism and ability to render the visual applications most important to today’s consumers — games, photos, movies, art, photoshop, YouTube, GoogleEarth, virtual worlds — the GPU rises to match the CPU’s “centrality” in the computing scheme.

Less obvious, the GPU’s attributes also make it useful for all sorts of non-consumer applications like seismic geographic imaging for energy exploration, high-end military systems, and even quantitative finance.

Perhaps the most exciting shift unleashed by the GPU, however, is in cloud computing. At the January 2009 Consumer Electronics Show in Las Vegas, AMD and a small but revolutionary start-up called LightStage/Otoy announced they are building the world’s fastest petaflops supercomputer at LightStage/Otoy’s Burbank, CA, offices. But this isn’t just any supercomputer. It’s based on GPUs, not CPUs. And it’s not just really, really fast. Designed for the Internet age, this “render cloud” will enable real-time photorealistic 3D gaming and virtual worlds across the Web. It will compress the power of the most advanced motion picture CGI (computer generated imaging) techniques, which can consume hours to render one movie frame and months to produce movie sequences, into real-time . . . and link this power to the wider world over the Net. 

Watch this space. The GPU story is big.

But somehow, The Economist was convinced that my research represents some “gloomy prophesy,” that I am “doom-mongering” about an Internet “overload” that could “crash” the Internet. Where does The Economist find any evidence for these silly charges?

In a series of reports, articles (here and here), and presentations around the globe — and in a long, detailed, nuanced, very pleasant interview with The Economist, in which I thought the reporter grasped the key points — I have consistently said the exaflood is an opportunity, an embarrassment of riches.

I’ve also said it will take a lot of investment in networks (both wired and wireless), data centers, and other cloud infrastructure to both drive and accommodate this exaflood. Some have questioned this rather mundane statement, but for the life of me I can’t figure out why they deny building this amazingly powerful global Internet might cost a good bit of money.

One critic of mine has said he thinks we might need to spend $5-10 billion on new Net infrastructure over the next five years. What? We already spend some $70 billion a year on all communications infrastructure in the U.S. with an ever greater portion of that going toward what we might consider the Net. Google invests more than $3 billion a year in its cloud infrastructure, Verizon is building a $25-billion fiber-to-the-home network, and AT&T is investing another $10 billion, just for starters. Over the last 10 years, the cable TV companies invested some $120 billion. And Microsoft just yesterday said its new cloud computing infrastructure will consist of 20 new “super data centers,” at $1 billion a piece.

I’m glad The Economist quoted my line that “today’s networks are not remotely prepared to handle this exaflood.” Which is absolutely, unambiguously, uncontroversially true. Can you get all the HD video you want over your broadband connection today? Do all your remote applications work as fast as you’d like? Is your mobile phone and Wi-Fi access as widespread and flawless as you’d like? Do videos or applications always work instantly, without ever a hint of buffer or delay? Are today’s metro switches prepared for a jump from voice-over-IP to widespread high-resolution video conferencing? No, not even close.

But as we add capacity and robustness to many of these access networks, usage and traffic will surge, and the bottlenecks will shift to other parts of the Net. Core, edge, metro, access, data center — the architecture of the Net is ever-changing, with technologies and upgrades and investment happening in different spots at varying pace. This is not a debate about whether the Internet will “crash.” It’s a discussion about how the Net will evolve and grow, about what its capabilities and architecture will be, and about how much it will cost and how we will govern it, but mostly about how much it will yield in new innovation and economic growth.

The Economist and the myriad bloggers, who everyday try to kill some phantom catastrophe theory I do not recognize, are engaging in the old and very tedious practice of setting up digital straw men, which they then heroically strike down with a bold punch of the delete button. Ignoring the real issues and the real debate doesn’t take much effort, nor much thought.

Lots of people have criticized my rough estimates of a couple hundred billion in new Net investment over the next five years, saying it’s closer to $5-10 billion, and I wonder what the heck they are thinking.

So, here we have Wu raising the specter of search engine bias and Lessig raising the specter of Google-as-panopticon. And this comes on top of groups like EPIC and CDT calling for more regulation of the online advertising marketplace in the name of protecting privacy.  Alarm bells must be going off at the Googleplex. But we all have reason to be concerned because greater regulation of Google would mean greater regulation of the entire code / application layer of the Net.  It’s bad enough that we likely have greater regulation of the infrastructure layer on the way thanks to Net neutrality mandates. We need to work hard to contain the damage of increased calls for government to get it’s hands all over every other layer of the Net.