Mad Dog 21/21: Goodbye Kitty
September 19, 2011 Hesh Wiener
You may remember the movie Apotheker Now, or perhaps the novel on which it was based, Bartz of Darkness. The story, as you undoubtedly recall, concerns Fulton J. Wintel, sent into the wilderness by the greedy tyrant Leopold 2.0 to bring back whacko Kemeny Kurtz. Wintel finds Kurtz, who soon expires, muttering, “The oracle, the oracle.” The tale portrays the underside of Silicon Valley during the Doofus Era, when companies dexterously developing and exploiting new technologies overwhelmed those that slipped even once, or looked like they might have.
It’s the whacko season. Hewlett-Packard and Yahoo are beset by turmoil in the executive suite. Microsoft and Intel are on the desperate defensive. Even IBM, which is about as different from Apple and Google and Amazon and Facebook as any technology company can be, is at risk, too, because its most profitable systems products are tortoises, and right now the hares are winning all the races.
The problems of the troubled players in Information Technology and the triumphs of the day’s victors are peculiar to each party, but beneath them may be found electronic gadgets the size of coins called systems on a chip, or SoCs for short. The smaller ones are about the size of a dime. The larger ones are as big as a quarter. An SoC is typically as functionally complete as a server motherboard because, small as it may be, that is what it actually is. A typical SoC will have one or more general purpose central processors, along with memory controllers, I/O processors, graphics engines, security engines, and very sophisticated systems management circuits that provide heat and power control technology.
The competition among SoCs occurs in three dimensions: Absolute power, bang for the buck, and bang for the watt. While the engines in larger scale servers and client devices are judged according to the same criteria, the technical and economic efficiencies offered by SoCs are vastly greater than anything available in computers larger than your hand.
SoCs are plentiful. The population of large and midrange servers is in the hundreds of thousands. The installed base of clients and small servers is in the hundreds of millions. But a census of SoCs that are humming away right now would reveal numbers in the billions.
Arguably the most noticeable applications of SoC devices are in smartphones like the iPhone, Android, and Blackberry and tablets such as the iPad. But there’s an SoC in every office or home network router, each firewall appliance and every smart set-top video box, high end TV sets and the still ephemeral GoogleTV. You can find them in printers, scanners, and fax machines. They run small medical devices. And there may be several of them in your car.
As small and often invisible as they may be, SoCs have quite a bit in common with their far larger forbears. An SoC will have an operating system. More often than not that OS will be derived from Linux, although some SoC software (notably iOS, and not to be confused with IBM’s former i5/OS for midrange servers or Cisco Systems‘ IOS for switches ) is in the BSD family, and Microsoft is hoping (praying?) it can make a big splash with its mobile Windows 8. RIM has a couple including QNX. Qualcomm has Brew. Nokia, or more accurately a foundation it backed, has Symbian. And there are more.
The most prominent architecture for the general-purpose computing elements in these devices is an ARM core, from ARM Holdings. ARM technology is the overwhelming favorite of phone makers. ARM doesn’t make or sell chips; it develops architectures that it licenses to others. After ARM the next most popular SoC CPU family comes from MIPS. Old-timers may remember MIPS as the brain building part of Silicon Graphics; youngsters may recognize MIPS as the platform for DD-WRT (a Linux derivative) first used with Linksys routers. The SoC market has other entries, too, including Sparc, and there’s room for more. For now, it’s a RISC world, but that could change at any time, particularly if Intel, AMD, or VIA can create X64 CPUs with power that is measured in milliwatts rather than watts.
Chip industry chatter suggests that SoC architectures and X64 technology will collide as fabrication technology shrinks to 22 nanometer or smaller geometries. Observers optimistic about Intel technology expect 22 nanometer “Ivy Bridge” chips due next year to be much more powerful than rival ARM or MIPS chips and much less power hungry than today’s skinniest X64 circuits. At the same time, the SoC makers will be increasing the complexity and power consumption of their top end offerings, and a byproduct of this will likely be a rise in their energy requirements.
Overlap, when it comes, probably won’t be for another year or two, by which time ARM will have rolled out its Cortex-A15 40-bit technology, a significant upgrade from its present augmented 32-bit A9 designs, and MIPS will have had a chance to pack a lot more computing power into its SoC offerings.
The most visible battleground is expected to be where tablets compete with skinny laptops (and netbooks, if that type of computer remains a living category). There is bound to be a similar battle at the low end of the server market, too, where the lower cost and improved power of SoC-based boxes could seed a high volume business. Microsoft has to be thinking about this, because Apple or Google could reimagine the home server in a way that quickly inspires a high volume, big aggregate bucks market, one that gives architectures other than X64 market cred. All of Microsoft’s big customers, the companies that make PC and X64 servers, are designing SoC-based devices. They are not confining their use of SoCs to smartphones or tablets. Pretty much every engineer in the computer business has seen the stunning little video of Glowball put out by Nvidia to promote the Tegra 3 quad-core SoC it is promoting for mobile clients. Nvidia refers to its Tegra 3 as “Kal-El,” which is a reference to the back story of Superman, and manages to keep adults, including engineers working for rival chip development firms, from snickering by adding that the Kal-El has a dozen GPU cores to help out its four ARM engines.
Nvidia is fighting it out with Qualcomm‘s Snapdragon SoC family, Texas Instruments OMAP 4 and forthcoming OMAP 5 products, Samsung‘s proprietary group of ARM-based SoCs, Marvell‘s Armada line, and others. The device makers are all trying to win the customers created by Apple with its ARM-based SoCs, chips that appear only in Apple products and that are, at the moment, fabbed by Samsung. But within Apple there are skirmishes between the ARM developers and the X64 crowd that makes Mac desktop and laptop machines.
Every time Apple announces new laptops that are powered by Intel chips, the Wintel world feels it has gotten another stay of execution. This is true even though so far there are no smartphones, tablets, or other SoC-based devices that can do multitasking, run productivity applications, support photo and video editing, or generally do computing the way X64 machines do. Real computing horsepower may yet come to SoC country, but not for this year’s Christmas season and, chances are, not for next year’s either. That could all change if Microsoft decides to build a rich version of Office for ARM or MIPS. Right now, users cannot even get Outlook for iOS or Android (but it might be on the way), and users of mobile devices that want to work with Exchange servers have to buy third party software. (Microsoft Exchange does try to provide support for iPhone and email users with Android devices can get TouchDown and other Exchange client software but not from Microsoft.)
Here is one reason computer makers are pretty nervous right now: Melon Meter. Melon Meter is an iPhone app that analyzes the sound you get when you rap on a watermelon with your knuckles. You hold the phone by the melon, tap the phone to get things started, rap on the fruit and bingo the software analyzes the sound to figure out whether the melon is ripe or not. Try that with your IBM server! Try that with your Dell or HP PC!
Melon Meter is one of a zillion examples of the way a thin client powered by a tiny SoC knows what’s going on around it and can turn what it senses into useful (and in this case entertaining) information. A smartphone can include accelerometers, magnetometers, light sensors, sound sensors, a proximity sensor (so it knows whether it’s near your ear or not), system sensors (keeping in touch with the voltage and temperature of the phone’s battery, for instance), a GPS navigation receiver, timers, modems for voice telephony as well as data telephony, low power radio transceivers (for WiFi, Bluetooth and, soon, near field payment authorization), and of course a touch- and gesture-sensitive screen.
That’s just the client getting data from its immediate surroundings. Smartphone apps get data from servers, too. Weather apps do periodic checks for storm alerts. Travel apps send reminders about departure times and other matters. Calendar apps send notices about appointments. Mapping and navigation apps mash the client’s localization data with data about terrain, roads, highways, structures, businesses and other institutional residents. And as if that isn’t enough, there are social information pooling servers that let your smartphone see what your friends or associates are doing or what others are saying about a locale or the weather or your last tweet or just about anything else that can grab your attention (or somebody’s attention, if you’ve finally run out of patience with bee swarm behavior).
We all know that any kind of server can do any of the things smartphones love or any of the bean counting inside legacy applications, or anything else that’s computational, too. Just how much of each a server can do depends on its power, its loading, its communications bandwidth, the data it has on hand, and where it is allowed to go seeking assistance. So, in theory at least, your IBM i could do whatever Yahoo does for you, your HP Itanic system could be programmed to notify your poker buddies that you’re going to be a half hour late, the blob of processing you rent from Amazon could help you watch Casablanca while you’re waiting for a train, and so forth.
But in practice you are going to use the computers that are easiest to use. And what is easiest will depend on what apps you have on hand, what you know about using the apps, what kind of client you are packing, what your network is able to carry with speed and ease, and, if you haven’t put the pedal to the metal with some kind of all-you-can-eat service plan, what things cost.
What seems unlikely is that you will be using your SoC-based client with your institutional IT systems, unless you are running a TN5250 or TN3270 app or using your phone along with an RDP package (I like 2X with its excellent Dumbo mouse) to reach a Windows gateway. Even enterprises that have built mobile-friendly Websites for customers rarely do so for employees. But that might change.
Business users realize that many if not most of the notebook machines their field personnel use can be thinner these days and still pack enough wallop to be effective. In some cases, companies have found that even a tablet is enough of a computer to support an end user on the road, but that is the case mainly when the client device is used almost entirely for email or with applications that do all the heavy lifting at the server end of the wire. Still, during the next few months quad-core SoC chips will give tablets and ultralight notebooks (and, if they catch on, dockable smartphones) lots more pep than they have right now.
Making it all the more dynamic, Intel and Advanced Micro Devices are scrambling to keep the mobile end of the computer business from escaping. Whether their offerings are chipsets or SoCs, very portable computers with batteries that last all day and then some are going to be getting quite a lot of promotion. But that may not be enough.
If the iPad remains the monster of the tablet world, the giants of the server business are going to have to do a lot of adjusting. In order to support clients that are a lot skinnier and weaker than netbooks, let alone laptops, servers will have to do a lot more work. Also, they will have to have very fast reflexes. Mobile client users are incredibly impatient. If an app doesn’t yield results instantly, it’s not going to get a lot of use.
Companies that can make money by supporting thin client apps are going to migrate to server systems that can deliver a superior customer experience at an affordable price. If the servers can be brought into the glass house, that’s what most companies would prefer. If they cannot, it will be hard for IT departments to defend their empires against cloud computing.
Can IBM and HP and Dell offer attractive solutions and still protect their profitable legacy server businesses? IBM has been able to preserve its server business and grow its services business at the same time. HP and Dell talk as if they can do it but in fact their server-related sales are so much smaller than IBM’s that it is really not possible to equate them. (Add the systems software and services in–that’s IBM’s server business.)
Oracle with its Sun hardware may actually be in the best position of all because at this point it has a lot to gain and very little to lose. If it makes its number one goal dropping the cost of a thin client transaction by a factor of five or ten, it could provide the shock and awe the legacy server outfits so desperately need. What Oracle would have to develop, though, is a strategy that lets it sell more stuff while giving customers the ability to run many, many more transactions. In other words, Oracle would have to sell systems that do more or less what the clouds built by Google and Amazon do, which is different from what the glass house systems built by legacy hardware vendors (including Oracle/Sun) do.
What it might come down to, at least in Wintel’s worst nightmare, is a market moving to systems tuned to talk nicely to Android and iOS, and for which that talk is disruptively cheap. And if these systems don’t make legacy technology look its best, well, that’s happened before. In fact, it is the story of just about every significant achievement in computing.