Moore?s Law Will Hit a Wall for Mobile in Next Phase of Growth
- July 22, 2010
July 22, 2010 ? I want to address a topic on technology trends we?ve thought about at Skyfire for a while now. In short, we believe the consumer experience on smartphones has improved exponentially over the last few years, but it may have hit a friction point as we enter the next phase of mobile growth.
We in the technology industry are all used to citing the relentless exponential progress of ?Moore?s Law? for desktop computing. But in smartphones, there are some special factors that are limits to growth, unless very novel solutions emerge. Moore?s Law is facing limits in cellular network capacity and mobile battery life, relative to the demands on the mobile system.
On a component level, devices will continue to improve, but we?re talking about smartphones and networks as a system. Contrary to conventional wisdom, once you factor in network capacity needs and battery size constraint, we don?t think the smartphone user experience will be able to catch up fully with the ever-increasing capabilities of computers, which enjoy fixed-line Internet connections?at least not in the next decade.
Conventional wisdom has held that Moore?s Law, named for Intel co-founder Gordon Moore and his prediction in a 1965 paper, would apply to all parts of digital electronic devices. Moore predicted that integrated circuits would double in capacity at lower cost every two years, and that has proved true. Most technologists now cite Moore?s law as applicable universally in technology for memory, processing speed, and even pixels in digital cameras. It?s shorthand for how fast technology moves. And those who compare an iPhone or an HTC Evo to what ?smartphones? looked like before 2007 certainly have seen incredible advances.
Moore?s Law has become so ingrained that it?s admittedly controversial to suggest smartphoneswon?t catch up with desktop computers and the fixed broadband Internet.
We believe that if you look at mobile subsystems as standalone components, then Moore?s Law will hold. But when combined as a system ? with RF towers and backhauland bandwidth requirements and display and processing and battery ? then we?re facing serious limits to growth.
So why would mobile be different than desktop computing?
The first big problem is network capacity. Although networks get better, what matters to users is the download speed relative to typical use cases, and throughput during peak daytime hours. Despite major advancements in smartphones and 4G network speeds, the explosion in mobile video traffic of nearly 6000 percent (per the latest Cisco Visual Networking Index study) over the next three years will strike as a tsunami crashing on the networks. This will create what I term ?Mobile Warming?, the severe strain on mobile networks that could mean slow-loading content, dropped calls, and frequently-buffering video.
While in theory, 4G networks will improve by 4x-6x network speeds, in practice a network is measured by its capacity to handle increasing traffic. Video traffic, as noted above, will rise 59x, according to Cisco, and overall wireless data to 39x. That?s a mismatch. The latest broadband-to-home services promise speeds approaching 100 megabits per second, and developers will build applications to use that. Desktop users can look forward to HD video conferencing, HD telemedicine, 3D gaming on webkit browsers, and other bandwidth-intensive applications that just won?t port well to cellular networks.
4G speeds are touted between four and 12 megabits per second. But once more users sign up for 4G, and are farther from towers and sitting inside thick building walls, speeds will run well below that, most RF experts privately admit. When users stray from core coverage areas and find themselves on a 2G EDGE or 1x connection, they will find speeds closer to 100 kilobits per second today. That is 0.1 percent the bandwidth of the latest home broadband. In real life, mobile smartphone users will find their experience a long way from what they experience when connected via fixed line broadband.
John Donovan, the CTO of AT&T, predicted at MobileBeat last week that 90 percent of consumer traffic on AT&T?s mobile IP network would be from video by 2014. Video involves an order of magnitude more data than any other use case.
Paul Jacobs, the CEO of famed mobile tech leader Qualcomm, said at CTIA last fall that engineers have squeezed with 4G technology nearly everything possible from the limited radio spectrum for cellular networks. He called for the FCC to open up more spectrum via auctions. That, however, is probably a minimum seven year process before it could be concluded, built out, and new networks (and brand new matching devices) could go live. This June, speaking at the Uplinqconference, Jacobs called on developers to be responsible: ?Total bits consumed by data traffic now exceeds voice traffic worldwide, but there?s only so much spectrum,? Jacobs said. ?You guys have an obligation to build applications that are as network efficient as possible. We all have a responsibility to be as efficient as we can.?
The trouble is that this is a classic ?tragedy of the commons? collective action problem. Any given app or developer or brand has little incentive to protect the network or sacrifice functionality ? they don?t bear the expense. Some publishers will mobile optimize, others may not care, or may not have the resources. That means network owners need to defend the network against bandwidth hogs.
This problem will only partly be abated by data quotas and tiers being rolled out by AT&T, which are sure to be followed by other operators. The number of users maxing their quotas, and trading up to bigger plans, will keep increasing.
As Adobe and Android gear up to ship full native Flash 10.1 on a few superphones with 1 Ghz processors, a whole world of video content will be opened up to mobile browsing that is not optimized for mobile. If you complain it?s hard to get a good connection now, wait until the user at the next caf? table is downloading a 720p HD Flash video on the same 3G network tower you?re sharing.
Apple and AT&T understood this constraint when Apple launched FaceTime, making video calling enabled only for WiFi. Imagine what video calling on the iPhone would do to AT&T?s fragile network.
And in the race for networks to get ready, more compression solutions are needed, to keep what we call ?mobile pollution? off the wireless networks. When a mobile user clicks on an HDor high quality video, the network is forced to download a mountain of pixels that the receiving smartphone can?t even display on a small screen, and at a frame rate that just won?t fit on a cellular connection. Most web video is intended for much bigger screens, for users on fixed broadband connections. Users are now getting the capability to ?break out? of very limited ?mobile optimized? sites and surf the real web, with all its fat multimedia. What is needed, are network-side or server-side solutions to protect the networks from pollution by compressing it, and mobile optimizing on the fly.
4G is starting to roll out, yet its coverage won?t be truly nationwide in the US for a few years. Verizon?s CEO Lowell McAdamsaid publicly in May 2010 that Verizon was looking for technologies to compress video on 4G LTE networks.One has to expect, as the number of users and devices on 4G grows, that 4G systems will start to get more clogged, as 3G and 2G networks have. That is, unless compression and filtering solutions are found.
The second big problem is that Moore?s Law works wonderfully when you can plug devices into a power outlet. More powerful CPUs keep coming on desktops and in servers, but they eat more power. And when you need a lightweight device that fits in a pocket, a handset maker is constrained on battery size.
Unless there?s a major breakthrough in battery technology, we just can?t fit much more power into the same size battery; and bigger CPUs require more power. Electrical ?leakage? occurs as battery transistors are shrunk beyond miniature size, meaning batteries can?t keep getting more microscopic as chips have. That?s why I find on the various Android phonesI test that they have trouble making it through a workday without charging. This is especially as we jump from 600 mhz to 1ghz processors on the latest ?superphones.? If I do an hour of calls, check email a couple times and browse for a couple minutes, I get the red low-battery signal. That is before the demands of video, gaming, video calling, and more. All of this battery sucking will bring power management to the forefront of design priorities. This will have trade-offs resulting in larger processors.
Because of battery life, smartphone makers can?t keep throwing in bigger and bigger processors to try to catch up with the processing power of desktop computers. Unless users are willing to charge their phones every two hours ? which would defeat the point of mobility ? the jumps in CPU power each year on smartphones will have to slow down. Alternately, users will have to refrain from using their smartphones as intensely as their laptops.
Cooling makes for another factor. More transistors on a chip generate more heat. On a desktop, you add a fan. When the phone gets burning hot in a user?s hand, where would you put a fan? It?s a real issue in a small form factor.
If smartphones can?t catch up with the latest desktop-equivalent chips then they will have trouble keeping up with the most computationally-intensive tasks that are routine on desktops, like playing un-optimized Flash videos. Whenever a smartphone encounters a video in a codec that does not match the hardware-decoding configuration on the phone?s graphics processor (GPU), or even video in the wrong screen ratio, the GPU gives up and leaves playback to the main CPU.
For instance, on a desktop computer, handling a new codec like Google?s VP8 (christened WebM) would be simple. On a smartphone, where VP8 is not present on any GPUs shipping or yet announced, VP8 video will have to be decoded in software. Battery life will be significantly affected whenever you decode video in software on the CPU, running the processor at 100 percent capacity. Over time, we expect Google will convince top-of-the-line chip makers to add VP8 support and upgrade software to allow VP8 decoding in hardware. My point is, for any given example a solution might be a couple years away on the most cutting-edge devices. But it takes years to roll out widely, and there?s always some challenge.
Another Intel engineer, Randall Kennedy, coined the term ?bloat? to describe how Microsoft Office in 2007 ran no faster than in 2000. The greater processing power from newer chips was eaten by more demanding software. The web keeps getting more and more complex, more ?bloated,? with HD video and 3D animations. Smartphones will strain to keep up on these cutting-edge demands, even if they seem fine on basic text and photo websites.
The two problems?network bandwidth and battery life?get magnified by video. Decoding video that is not optimized for mobile will burn battery at incredible rates. The vast majority of smartphones today have hardware decoding only for limited codecs like h.264 ?baseline? profile, but not for other flavors of h.264 such as ?main profile? or ?extended profile?. When users of the latest superphones encounter a Flash video in the wrong flavor, they will find choppy performance and fast battery burn.
This fragmentation is also why it?s a myth so far that HTML5 video tags will prove any better at solving the mobile video problem than Flash. HTML5 rightly has many proponents, Skyfire included. Yet the battle over what the HTML5 standards will be is just beginning. Like Flash, the HTML5 video tag is really just a container. Publishers can put video in with any video codec and streaming protocol they want. There are no universal standards today. Apple has favored h.264 and its own flavor of HTTP streaming protocol. Google now pushes VP8 (webM). Microsoft favors Silverlight and IIS streaming protocol. Mozilla has categorically refused to enable h.264, due to rights issues and fear of licensing fees, and now plans to endorse VP8 or OggTheora. What?s a publisher or handset manufacturer to do?
It?s very expensive to build and maintain separate sites for each mobile and OS ecosystem. Fragmentation is proving very persistent, which makes optimizing battery life even harder. Anytime the video codec, screen resolution, and bit rate is not tuned adaptively to network conditions and device capabilities, then performance and battery life suffer.
Clouds Help Avoid Mobile Warming
Our company, Skyfire, believes that solutions will be found in cloud computing, among other new technologies. Skyfire itself aims to tackle these issues with its cloud-computing Rocket platform, which enhances mobile video browsing by compressing video by 75 percent on average, in comparison to native Flash on high-end smartphones. Skyfire?s real-time transcoding also pre-processes video to translate it into the right format for each device type, enabling 100 percent hardware decoding of video, which can boost battery life by over 30 percent for users who play a lot of video.This means less buffering, faster video load times and smooth video playback. Essentially, Skyfire uses the cloud to normalize the messy fragmentation of the web, and give the best user experience for each user in adaptive real time.
Summing Up
In the end, Moore?s Law at the component level probably will continue to hold true. We are all indebted to Gordon Moore?s insight. But as a system, I?d argue that mobile will face special constraints that will keep a material gap with the desktop Internet, unless special solutions are deployed on the server side.
- Jeff