How Microsoft solves touch on the desktop

Here at ExtremeTech we’ve talked fairly extensively about Windows 8, particularly about how the changes coming to the operating system will affect longtime Windows users. This certainly remains a concern, and the Building Windows 8 blog has been keeping us informed with semi-regular, gargantuan epistles, but I’ve still been thinking a lot about how normal people will interact with Windows 8. These cubicle dwellers, students, and casual users won’t be on tablets and they won’t know whatWindows RT is, but soon many of them will be staring at a tile-tastic Start Screen, wondering that happened to Windows.

Microsoft seems more than comfortable with the disruption — which is refreshing to see from the company — but it will still need to give desktop users tools with which the new OS can be operated. Originally it seemed like all the touch controls and new conventions would be additive and desktop users could just ignore them if they wanted, but that might not have to be the case. With the $40 Microsoft Touch mouse users without a touch display will be able to swipe, slide, and pinch their way through Windows 8.

Microsoft’s Touch mouse has been available for the better part of a year and touch gestures in Windows are nothing new, but we know that they will play a prominent role in Windows in the future. This mouse, and ones like it, could bring multitouch controls to the desktop-using masses, making them both affordable and accessible. It could be one of the tools that let’s Windows 8′s designers achieve both their user experience and HCI goals.

Of course the mouse will work with laptops, but trackpads will be a lot more convenient and popular option. With laptops outselling desktops these days there is no question that the trackpad will be the primary gesture tool in Windows 8. And the Touch mouse will never be for Microsoft what the Magic Trackpad is for Apple — the canonical device by which Apple’s laptop and mobile device interactions are shifted to the desktop — but it could still play an important part in allowing desktop users to take advantage of the features that Windows 8 offers.

With the mandate that Windows 8 “work[s] like a device, not a computer” it’s clear that Microsoft is banking on products like this one to bridge the gap between different classes of users and different modes of Windows usage. If you think that’s strong language, thecompany also noted that it wants “touch as a first-class input method” and that it’s “embracing” touch on the desktop. How do you do that without a refresh to most of the displays in the world and a some nasty cases of “gorilla arm“? A capacitive, multitouch-capable mouse of course.

The final piece to this puzzle is Microsoft’s announcement in February that the Touch mouse will be updated for Windows 8. Specific changes were not mentioned, but I’ve been told that there will be more gestures and better gesture compatibility once Windows 8 hits GA status. Swipes will be able to handle tasks like summoning Windows 8′s charms and app bar, as well as switching between programs. Nothing you can’t with the keyboard, but with the focus on touch it’s important to give desktop users an option.

Source

Tech wrecks — Lessons from some of the biggest hardware screw-ups

Tech wrecks often headline our daily news, mostly the result of operator error or equipment failure — generally they are “one-off” incidents which leave the front pages as quickly as they arrive. Some failures, though, start with faulty designs, and are doomed to be repeated until diagnosed and fixed. In addition to making for good headlines, these sobering bits of high-tech history offer some valuable lessons that can help shape the way designers think and lawmakers regulate.

We’ve rounded up some of history’s most serious, famous, and instructive hardware design failures for you — avoiding bridges and buildings, as they really need an article or two of their own. Some of these failures were fatal, some a nuisance, and some even a bit humorous. All have lessons for anyone involved in designing or developing high-tech products.

Fire in the hold — Sony laptop batteries

Typical of many design failures, overheating and fires caused by manufacturing defects in Sony-made laptop batteries in 2006 were met with corporate foot-dragging and denial. After Dell recalled over four million of the batteries, Sony insisted Dell was the only computer maker affected.

A week later, after Apple had to recall millions of batteries, Sony changed its tune to claim the damage stopped with Apple. Other laptop vendors including Lenovo and HP were quick to echo the sentiment, giving technical explanations “proving” that they wouldn’t be affected. Yet, over the next few weeks nearly every major laptop maker — including Lenovo and HP — had to recall some of the Sony batteries they had been selling. Sony, and apparently Dell, had known about the manufacturing issue that caused the problem — issues in fabrication that left bits of metal in the cells — many months earlier, but decided the problem wasn’t worth fixing until the fires began being reported by customers.

PC Pitstop recreated the conditions for an exploding laptop battery in their labs, which makes for this compelling demonstration:

It’s easy to forget how widespread the effects of the battery problems were. This news report enumerates the wide variety of issues it caused, like fires in planes and burned vehicles:

The far-ranging impact of the faulty batteries is a great illustration of the law of unanticipated consequences. During the height of the controversy, the number of actual fires reported was well under 100 — a minuscule fraction of the tens of millions of laptops sold. However, when those fires were in cargo holds or in vehicles, the otherwise small failure of a battery and perhaps loss of a computer was magnified into a potentially major disaster.

Another unanticipated consequence was the firestorm of publicity that necessitated the recall. Sony learned that even a few product safety incidents were worth taking seriously, and changed its defect reporting policies after the battery debacle — a lesson Toyota would need to relearn in spades when reports of their vehicles accelerating on their own started to come in.

Square windows that changed the aviation industry — de Havilland Comet

Imagine an aviation industry led by Britain’s de Havilland, with Boeing in a distance second place. That’s how it was after the Comet — the world’s first jetliner — was introduced by de Havilland in 1952. Sleek and fast, it eventually cut six hours off the flight time from New York to London.

In order to take full advantage of its higher-powered jet engines, the airplane flew at 35,000 feet — providing a faster and smoother flight. High-altitude flying is standard today, but pressurized cabins were new and poorly understood at that time. The Comet’s designers opted for large, squared-off windows because they looked more attractive than the simpler, round “porthole” style that had been more traditional. Unfortunately for the Comet, and for dozens of passengers who would die in several resulting crashes, metal fatigue was also not well understood. Stresses piled up around the square corners of the windows, and over time planes began to drop from the sky.

Since the crashes were at high altitude and often over water, it took time before the problem could be traced to the size and shape of the windows. The sleuthing wasn’t successful until after a full-size fuselage was repeatedly pressurized in a water tank — and it failed near the window corners. Once the flaw was uncovered, the fleet was pulled out of service. While the Comet was being redesigned with new windows and a thicker skin, Boeing’s 707 and Douglas’s DC-8 were introduced and became airline favorites. By the time the newly designed Comet 4 re-entered commercial service in 1958 it was too late. Primacy in commercial aviation had moved from the UK to the US, never to return.

Engineers from both Boeing and Douglas are reported to have told de Havilland privately that they also had no idea about the fatigue and pressurization problems, and may well have made the same mistake if it hadn’t been for the Comet. Design problems were also magnified by the manufacturing process. Rivets attaching the windows were punched instead of drilled, creating more stress than expected by the designers — who had planned for drilled rivets. Wherever the blame is laid, the aviation industry benefited from the intensive analysis and resulting learnings about pressurization and metal fatigue.

Faulty O-ring — Space shuttle Challenger

Like most design-related disasters, the catastrophic failure of a pair of O-rings in the space shuttle Challenger was the result of more than just a poor design by Morton Thiokol, contractors for the ship’s solid rocket booster (SRB). NASA was also implicated by the post-mortem Rogers Commission investigation for ignoring warnings about the design for nearly a decade. A launch in cool temperatures — also flagged as a problem – exacerbated the defect, and the result was one of the worst space disasters in history.

78 seconds after launch, an O-ring (and its backup) in one of the SRBs failed, allowing gas to vent outside and begin to rip apart the shuttle. Without any escape system, all seven crew members were doomed once the craft separated into pieces. While the inadequate dual O-ring design (one primary and one backup) is certainly to blame, the incident is also a case study in how organizations ignore warning signs at their peril.

At various times over the nine years prior to the Challenger’s launch, engineers at both NASA and Morton Thiokol realized there were problems and reported them. Even right before the launch, Thiokol recommended a delay due to concerns the O-rings would fail in the cold weather. At that point it was NASA that urged the launch not be postponed — although they didn’t know the whole story about the previous concerns of design engineers about the O-rings. The seal around the O-Rings was being redesigned as the Challenger mission took flight, but the problem was never deemed serious enough to ground the shuttle program — until after the Challenger explosion, when it was halted for 32 months. When the shuttles resumed, an improved triple O-ring design was in place.

Terror on ice — Olympic bobsled run

Greed is at the heart of many tech-related tragedies, and the accidents at Whistler’s luge run built for the Vancouver Olympics were no exception. Hoping to extend the life of the run to gather tourist dollars after the big event, the track was situated on a cramped spot at the popular Whistler ski resort, instead of on a larger, safer space near the city. Normally, sled runs are carefully designed to provide exciting competition at high speeds while still providing for the safety of the competitors. Like any other design challenge, though, eventually the performance envelope gets pushed past the boundaries of safety — in this case magnified by a narrow and steep location for the course.

Even after record-breaking runs of nearly 100 miles per hour and a resulting fatality caused the Luge federation to restrict future courses to 87 mph designs, they also decided not to try and fix the Whistler track. At least by then many minor changes had been made to the run’s safety walls and the starting location to enhance safety.

As we’ve seen with our other design disasters, there was a large paper trail of concerns, legal opinions, and denial leading up to the fatal accident — which only came to light once someone was killed. The Olympic organizing committee (VANOC), the International Luge Federation (ILF) and the run’s designer, Udo Gurgel, took turns both denying they knew of problems and revealing that they had long had concerns. None of them, or the athletes — who had nicknamed turn 13 “50/50″ for their chances of making it through on course — had the courage to actually stop the event or make what proved to be important changes until international attention was drawn to the course by the high-profile accident right before the opening ceremonies of the 2010 Winter Olympics.

While not a conventional hardware failure, as we’d normally think of it, this was a failure of design and a failure to take preventative measures into account. In other words, it might seem different because of the setting and involvement of sport, but it has all the markings of a tech wreck.

First computer bug — Hopper’s Aiken Relay Calculator

When we think of bugs today, they’re usually a software problem. This poor moth, though, was found stuck in the early Aiken Relay Calculator that Grace Hopper’s team was trying to program in 1947. While the moth’s namesakes — computer bugs — have been responsible for hundreds of disasters of all sizes, and likely thousands of deaths, this poor critter did nothing more than cause the group to waste some time and it gave up its life for post-mortem glory. Amazingly, the text “First actual case of bug being found” was actually written in their log by the operators who found the moth. How did they know that bugs and debugging would go on to become two of the most famous terms in computing?

That’s actually the more interesting part of the story — the real history of the term bug. It turns out that there was an antecedent for the use of the word that explains the log entry. Edison, decades earlier, had described the process of troubleshooting a technical glitch in his phonograph as looking for an (imaginary in this case) bug. Even Edison may have borrowed the term from telegraphers — Edison got his start in telegraphy — who used it to describe problems possibly due to insects getting into the cables. So it makes sense that Hopper’s team, with a sense of humor, would refer to the moth as an “actual case” of a bug being found. Perhaps the lesson here is that you don’t have to be the first to use a term to become the one famous for it. Probably no consolation to the now infamous moth.

Purists will point out that the bug isn’t really a technical flaw, but I wanted to end the article on a humorous note because, frankly, failed hardware usually isn’t all that funny. And in any case, the photo is too good to pass up.

[Image credits: Challenger explosion, Bug log, laptop fire, bobsled, Whistler Sliding Centre,computer]

Source

HDD Pricewatch: Higher prices are the new normal

It’s been several months since the flood waters in Thailand began receding and since ExtremeTech posted its last edition of HDD PriceWatch, but we’ve finally begun to see more stable prices for various hard drives. Through February and March, prices fluctuated significantly, often moving in contradictory directions. Now that things have settled, how’s the storage market faring?

For the big dogs, things are going pretty well. As we predicted, there’s been a wave of further consolidation; Hitachi and Western Digital have finalized the sale of the former’s HDD business unit to the latter, while Seagate recently announced its intended purchase of rival LaCie.

Way back at the beginning of this mess, we selected a basket of drives by manufacturer, capacity, form factor, and performance. We’ve had to make some alterations along the way, but have managed to keep things fairly consistent. Here is what we’ve uncovered…

We started off tracking the Western Digital WD10EARS but that drive has apparently been deprecated in favor of the EARX. Performance between the two should be identical, but while prices have come down from January, they remain far above Q3 2011 levels. That trend holds true across the board; prices are generally down from where they were, but nowhere near where they used to be.

Here’s the percent above baseline data in chart form. Again, solid declines are visible nearly everywhere; Western Digital’s price increases on its 2TB Caviar Green and Caviar Black are outliers. When we spoke to Dynamite Data’s Kris Kubicki last winter, he predicted that we’d see manufacturers dropping SKUs and reducing the number of legacy drives they built, and that may be happening. Several lower capacity drives have disappeared, including Samsung’s Spinpoint MP4. The good news is, the 750GB M8 HN offers more than twice the capacity for the same $89 price point the M4 was hitting in January.

Are things going to change any time soon? We doubt it. WD and Seagate both reported record profits this past quarter. In Q1 2011, Western Digital reported net profit of $146M against sales of $2.3B while Seagate recorded $2.7B in revenue and $93 million in net income. That’s a net profit margin of 6% and 3%, respectively. For this past quarter, Western Digital reported sales of $3B (thanks in part to its acquisition of Hitachi) and a net income of $483 million, while Seagate hit $4.4B in revenue and $1.1B in profits. Net margin was 16% and 37% respectively.

With profit margins like this, the hard drive manufacturers are going to be loathe to cut prices. After years of barely making profits, the Thailand floods are the best excuse ever to drive record income for a few quarters. All of this means that while we expect prices to gradually decline, holding off on a necessary purchase doesn’t make much sense. If you need a drive, you need a drive; another six months may not show a dramatic return to form.

What About SSDs?

One of the predictions in the wake of the HDD floods was that SSD prices and adoption would spike thanks to the shrinking cost difference. That spike never materialized — SSD prices have continued a steady downward trend. Drives with full-featured controllers that would’ve been considered high-end just a year ago have now dropped to approximately $1 per GB.

128GB is an important price/capacity ratio for SSDs because it’s the point at which the drives offer enough storage space that switching over doesn’t feel like trying to slip into pants you haven’t worn since Bill Clinton was president. So how do these prices compare to hard drives in terms of cost-per-GB?

The hard drives still have a clear advantage there, despite the higher prices. Even Western Digital’s VelociRaptors, while still significantly more expensive than other hard drives, are just half the per-GB ratio of the SSDs. Then again, the shrinking gap points to the VelociRaptor’s relatively limited lifespan. The only way for WD to push the access speed envelope is to start incorporating cache in a hybrid structure like Seagate’s Momentus, increase the drive’s spindle speed to something approaching 15K, or both. The former seems far more likely than the latter; 15K hard drives run hotter, draw more power, and are noisier than anything enthusiasts are used to dealing with.

We don’t expect Flash prices to keep falling at the present rate; there’s a mountain of evidence to suggest that process node scaling will step on the breaks on that process. HDD capacities, in contrast, are headed into the stratosphere. Rather than slamming together in epic confrontation, the two standards are likely to co-exist for quite some time, with SSDs reserved for higher-end systems and a mixture of external drives, standard HDDs, and possibly HHD hybrid solutions picking up the majority of the market. That’s not to say we won’t see more notebooks continue moving towards solid-state storage at lower price points, but the fundamental capacity-vs-speed tradeoff isn’t going away.

We’d like to thank Kris Kubicki and his data analytics firm Dynamite Data for providing additional price information and market trends.

Source

Testing out Axis, Yahoo’s search-centric browser

Yahoo is taking a baby step into the browser wars with the release of its own search-centric browser. Known as Axis, it comes in the form of a plugin for HTML5compliant browsers on the desktop and a full-fledged browser for the iPhone and iPad.

Two concepts take center stage in Axis: instant responses and visual search. As you type in a search term, Axis returns results instantaneously based on what it thinks you may be searching for. The results appear as thumbnails across the top of your screen in iOS, and across the bottom on the desktop. You can scroll back and forth through these results to find the result you are looking for.

Visual search done right

On iOS, swiping across the page takes you from one result to the next. This is a nice shortcut for those of us who like to quickly search through multiple sources. There is a similar shortcut on the desktop: move your mouse to the sides of the browser windows and navigation arrows will appear. The visual search component of Axis is its most compelling feature, and is much nicer to look at (and usually tells you more) than looking through a list of links and text.

Given that Axis is multi-platform, Yahoo baked in a feature that allows you to continue browsing between devices. The last page you access on a device is sent up into the cloud, and then displayed on the Axis home page. Click on it on the other device, and you can continue where you left off.

This is a great feature for those of us that move from device to device frequently, although some may not like the fact that Yahoo is inherently tracking our every move.

The visual search and multi-device browsing are the two most compelling reasons to give Axis a shot, but also consider here is that Axis really isn’t a browser in the true sense of the word: it’s built upon pre-existing ones. This is important because pages will render as you expect them to, and only affect how your interact with your browser and not the way it works.

On iOS, Axis is more like a browser, but is really Safari within a skin due to Apple’s stringent controls on apps. That said, based of the time I spent testing I’d argue Axis is a better browser for iOS users than Safari due to its extra features.

Microsoft wins too

While Yahoo might have scored a win here with Axis, it was a big day for Microsoft as well. Remember that Bing powers Yahoo’s search results. If Axis takes off on mobile, it will be cutting into a fairly sizable chunk of Google’s search share.

One of the reasons that Google has been able to maintain dominance is its near stranglehold on mobile search. The two major mobile platforms — iOS and Android — both default to Google when it comes to search. If these folks now turn to Yahoo’s Axis, obviously that traffic’s going somewhere else.

It’s going to be interesting to watch — Axis is already 23rd on the top apps in the App Store in just its first day of release with not much in the way of promotion.

Source

Five things I’ve learned from Diablo III

Who says educational games have to be boring? I’ve learned a lot more than I expected this week from Blizzard Entertainment’s release of Diablo III. Here are the top five tidbits…

1. PC gaming is alive and well

There’s a lot to complain about regarding how Diablo III‘s launch was handled (ahem, see the next item), but one thing it proves is that reports of the PC’s death as an original gaming platform remain greatly exaggerated. One reason so many people get down on running games on computers is because of how many games aren’t designed to actually run on them. They’re designed to run on consoles andcomputers, which isn’t remotely the same thing and, at least to me, isn’t remotely as much fun. (I “love” games like DiRT 3 that don’t have basic mouse support, or Batman: Arkham City, which gives you no way to change graphics detail settings in-game.)

But when a game like Diablo, which makes no attempt to hide its pro-PC bias, it’s always gratifying to see it become a success. It reinforces what, on some level, we all already know: Because everyone already has a PC of some sort, there’s no better way for everyone to play everything. This means that most (if not all) console gamers aren’t left out in the cold, either, they’ll just need to put down their specialized controllers for a while. That’s okay: Given what I’ve had to endure from bad console-to-computer translations over the years, I think they can swallow a taste of their own medicine. And who knows? They might even discover that PC gaming isn’t so bad after all.

2. Requiring an internet connection is doing the devil’s work

When a game has the kind of relationship between single- and multiplayer modes that Diablo III has, some problems are naturally going to arise. But is forcing the player to have a persistent internet connection the only way to solve them? I’m not convinced. Blizzard’s servers’ launch-day woes were well documented and criticized (as they should have been), but the problems don’t end there. A couple of days ago, I spent upwards of an hour mapping and clearing out an area only to unceremoniously booted because of some server problem. When I was finally able to get back into the game 15 or so minutes later, all my progress had been erased and I had to start over from scratch.

In the grand scale of tech tragedies, this is pretty far down there, I admit. But having your single-player experience ruined or, worse, denied because of something beyond your control is not fun — and not least because you’ve already paid for the game, and don’t you deserve to be able to play it whenever you want? I understand that Blizzard needs some method to protect gamers from griefers, but there must be a better way than this. How tough would it be to include a mode that lets people play just by themselves and not worry about the grander network? You could even still require them to log in while loading the game, and most people aren’t going to complain too much. But the idea that any hiccup anywhere along the line can cause you to lose progress and patience isn’t going to be most people’s idea of a good time. Forcing players to stay online all the time may solve some problems, but if it creates more — and engenders bad will along the way — how good a solution is it really?

3. You don’t always need an expensive video card

This is something I know instinctively, but a reminder is always useful (especially given how much I’ve been writing about video cards around here lately). Although first-person shooters and titles with intense full-screen action will always benefit from first-rank discrete hardware, you don’t always need it — and, in fact, major titles let you get by with a lot less. Diablo III is an excellent example, with exceedingly modest system requirements (and only slightly more severe recommended components). On lesser hardware, the game might struggle just the tiniest bit during attacks by particularly large zombie hordes, I saw well-above-average performance regardless of the video hardware I used, without having to futz with the detail settings that much.

A lot of developers could learn from Blizzard’s example, both on World of Warcraft and here: A big part of why their games are so popular is because pretty much anyone can play them. That’s smart thinking, from both a business and a technological standpoint.

4. Twelve years may be too long

Those who played Diablo II back in 2000 obviously haven’t forgotten it and were probably first in line (or, er, online) to playDiablo III, and their fanaticism is undoubtedly a big part of what’s generating excitement for new audiences today. But 12 years is a long time, especially on the developers’ side — long enough, in fact, to make a new game feel older than the one that preceded it.

I share many of the concerns in this Kotaku story, “The Dumbing Down of Diablo III?”. A semi-official refutation of its points showed up on Battle.net, and it’s pretty persuasive, but if Diablo III captures the basic look and atmosphere of the previous games, I have yet to be thoroughly entranced, in no small part because it doesn’t seem like I’m really allowed to be myself in the game. Is this really an enormous drawback given that I don’t typically fight brain-munching denizens of the underworld? No. But it circles back to the internet connection issue: I want to feel like I have some say in what’s going on, and someone else isn’t making all the decisions for me.

Personally, I find “customizing” my character with items and runes less compelling than “rolling” him from scratch. That’s me, however; general expectations are different today, and the people at Blizzard have to meet them if they want their game to sell. But is it possible that maybe, just maybe, the first two Diablo games had qualities that differentiated them from their competitors, whereasDiablo III is more concerned about fitting in? That’s what it seems like to me. There’s nothing inherently wrong with such a strategy, but it definitely points to the casual-ization of mainstream gaming, which I find more troubling than I do heartening. It shouldn’t be a bad thing for a game to look and play in a unique away, but most of those qualities have been rubbed out of Diablo III. Am I having a good time with it? Sure. But it’s strangely forgettable, even tame, and I’m seriously wondering whether I’ll still care about a month from now — let alone another 12 years.

5. Gaming doesn’t need to be a huge investment

If you’ve been following the Diablo series since its debut in 1996, chances are an army of undead couldn’t stop you from devouring the latest chapter. But if you’re new to the whole thing, and are coming to it mostly because you’ve gotten swept up in the hype, hold on to your credit card for just a moment and consider your options. If this is the kind of game you want, you can do about as well (give or take) for less money, and not miss out on much. One of the more consistently popular ways is Torchlight, another isometric 3D action RPG that you can download for just $15 on Steam. I’m not going to pretend that it’s as original or as thematically rich as Diablo, but the basic look, feel, and game play are all there, and you can save as much as $45 this way. (The sequel, Torchlight II, has just entered its closed beta test phase, and is slated to be released this summer for a mere $20.)

If you like the action but are either turned off by the graphics or if, despite what I said a couple of sections ago, are concerned that your computer can run them properly, try checking out the classics of the genre from which these contemporary titles sprang: Nethack and Angband. You’ll have to sacrifice a bit on the graphics side — okay, you’ll have to sacrifice a lot on the graphics side — but you’ll get tense, varied, and absurdly creative gameplay from these titles that dwarf anything Blizzard has devised. Want more of an incentive? They’re completely free. Be warned, though: These are about as addictive as PC gaming — whether of the high or ultra-low-tech variety — gets.

Source

How big is the cloud?

Last month, ExtremeTech revealed to youthe true scale of internet porn. At any one time, streaming adult videos probably utilize around 30% of the internet’s total bandwidth, which equates to around 6 terabytes of porn being consumed every second. But what about the other 70%? Netflix, YouTube, and other non-adult video sites are huge bandwidth hogs, possibly accounting for as much as 40% of internet traffic. Digital file lockers, such as Rapidshare and Megaupload, account for around 10% of traffic worldwide. Web surfing and email (and spam!) are another 15%. And then there’s cloud computing.

Today, the vast majority of web services and sites are hosted in the cloud. By this I mean that, instead of companies (such as Ziff Davis/ExtremeTech) managing their own hardware, third-party cloud storage and computing services are used. Amazon Web Services (AWS), Microsoft Azure, and Google are three prominent examples of huge cloud clusters, but there are hundreds of smaller operations that range in size from a whole data center down to a few racks.

The power of the cloud is vested in the fact that it can be coerced and shoehorned into tasks as disparate as a cloud-based supercomputer, to webmail, to simple document storage. On a single cloud cluster, Google can host and serve petabytes of YouTube videos and store all of your email and documents. Of all the facets of the cloud, though, today we’re going to focus on cloud storage.

A Microsoft data center

While storage might not be as sexy as terabytes of RAM and thousands of CPU cores, it is the most reliable way of measuring the size of the cloud, especially when we factor in bandwidth usage. From the total amount of storage we can also work out the cost of cloud storage — and from there, we can finally work out why the likes of Google, Microsoft, and Dropbox are falling over themselves to provide cloud storage services.

Like the porn story, we’ll first start with some theoretical numbers, and then move onto some real-world figures (and hardware) from Backblaze, a cloud backup provider.

Petabytes

For the most part, real numbers from the big companies, such as Google, Facebook, Amazon, and Microsoft, are few and far between. If you scour the web, though, some rough ballpark figures emerge:

• Facebook, in its IPO filing, said it stores over 100 petabytes (PB) of media (photos and videos). It’s not unrealistic to say that Facebook probably has a total storage of capacity well beyond that, once you factor in backups and other data (status updates, likes, and so on), possibly in the 300PB range.
• Microsoft recently admitted that Hotmail stores over 100 petabytes, and that SkyDrive, with “17 million customers,” stores 10PB of data. Like Facebook, Microsoft’s total capacity, once we factor in the rest of Azure and its web properties, is probably well over 300 petabytes.
• Megaupload is relatively tiny in comparison, apparently storing just 25 petabytes.
• Amazon, rather than giving us a nice, easy number of petabytes, instead announces the total number of objects stored by its S3 cloud storage service. As of April 2012, Amazon S3 stored 905 billion objects. If we assume an average size of 100KB, that’s around 90 petabytes; if the average size is 1MB, that’s 900 petabytes — almost an exabyte!
• Dropbox, a year ago, stored “10+ petabytes” of data. It had 25 million users then, and 100 million users today, so all things being equal the company now stores around 40PB of data.

To put these figures into perspective, an average computer probably has a 500GB or 1TB hard drive, and a petabyte is 1024TB. At the very least, then, Microsoft and Facebook data centers play host to more than 100,000 hard drives. Without building custom hardware, you can squeeze 48 drives into a 4U enclosure. After accounting for networking gear, that means you’re probably looking at around 400 hard drives per 40U rack — or 250 racks, each of which occupies around one square meter of floor space. This might sound like a lot, but when you consider that Google, Amazon, Facebook, and Microsoft regularly roll out data centers with floor plans of over 30,000 square meters (300,000+ square feet), it’s really not that much. In the grand scale of things, a lot more space is dedicated to servers (i.e. CPUs) and networking gear.

Bandwidth

Bandwidth-wise, we have even less data from the big boys. We know that, as of last year, one million files were being saved every five minutes — so today, with four times as many users, that’s 800,000 files per minute. Amazon S3, which is significantly larger than Dropbox, handles “650,000 requests per second.”

If we assume that the average file stored on Dropbox is 500KB (a mix of photos, videos, and documents) then Dropbox stores a total of 400,000 megabytes (0.4TB) per minute — or 6.7GB per second (54Gbps). We don’t have any data on how much data Dropbox sends per minute (i.e. people downloading files from their Dropbox), but it’s probably in the region of 10 to 20Gbps.

Amazon S3, which is mainly used to store static files for websites (images, style sheets, videos), probably has a lower average file size than Dropbox. If we assume an average size of 100KB per file, then 650,000 requests per second comes to a grand total of 61 gigabytes of data transferred per second, or 488Gbps. This is very close to the 800Gbps figure that we estimated for a large porn site, which equates to around 2% of total internet traffic — Amazon is pretty darn big!

Facebook and Microsoft, with between 100 and 300PB of storage each, probably fall somewhere between Dropbox and Amazon in terms of bandwidth usage — maybe 200Gbps a piece.

But enough theory! Let’s discuss some real-world numbers and real-world hardware!

Backblaze

Try as we might, every cloud storage provider rebuffed our requests for information on their hardware and software setups — except Backblaze, one of the few remaining unlimited cloud backup providers. Backblaze is relatively small, but the real numbers will still astound you.

First things first: Backblaze stores a total of 32 petabytes of user data, mostly fashioned out of 3TB hard drives. For redundancy, groups of 15 drives are banded together using RAID 6, which places two parity blocks on each of the 15 disks. RAID 6 can rebuild data from the parity blocks as long as there is no more than two concurrent disk failures in a group of 15 — and yes, replacing faulty drives is by far the most important task of whoever is on call at the Backblaze data center, so that redundancy can be restored as soon as possible.

Backblaze adds 3 petabytes to its storage cluster every month, in the form of 135-terabyte “pods” — custom-designed 4U rack mounted computer cases that accommodate 45 hard drives each (pictured below). 3 petabytes per month means that Backblaze is adding roughly 22 pods per month — which equates to almost two standard 48U racks.

While you might think that the pods themselves are a major operating cost (they cost $7,384 each), hardware is really the least of Backblaze’s concerns; power, data center space, and bandwidth are far more expensive over time.

At $0.20 per kilowatt hour (kWh) each pod costs around $100 per month. Data center power is incredibly expensive, which is why many web companies are building data centers in locations with cheap power, such as Oregon. Backblaze currently hosts its servers in a third-party data center, which means it has to pay for the floorspace occupied by its cabinets; again, this works out at around $100 per pod (~$1,000 per rack). At the time of writing, Backblaze has a total of around 300 pods stored in 40 rack cabinets — so around $30,000 is spent per month on electricity, and around $40,000 is spent on floor space.

And then, of course, there’s bandwidth. At any one time, Backblaze’s routers have 12Gbps of data inbound (about 1.5 gigabytes per second), and 500Mbps (62.5MB/s) outbound. The disparity is because Backblaze is a backup service — but even so, that 500Mbps figure would suggest that hard drives fail (or data is lost) much more than I would expect. At the moment, Backblaze receives around 100 million files from its users per day, or 36.5 billion files per year — and in 2011, it restored a total of 600 million files. Unfortunately we don’t know how exactly how many restored computers that equates to, but at a conservative estimate of 5,000 files per computer (excluding Windows and program files) that comes to 120,000 very relieved backuppers.

Back in its salad days, Backblaze paid $30 per Mbps per month — now, thankfully, that figure’s down to $2/Mbps/month, which comes to a total of around $25,000 per month. Backblaze does have cheaper offers on the table from other ISPs, and again the company would save considerable money if it built its own data center.

All told, then, the total operating cost (before employees!) of a 40-petabyte cloud storage cluster is $100,000 per month, split fairly evenly between floor space, bandwidth, and power consumption.

Why cloud storage is a cut-throat business

You may have noticed that, in recent months, Microsoft, Google, Dropbox, and others are fervently fighting to be your cloud storage provider of choice. At first, I couldn’t work this out — but now, after looking at the numbers, it becomes a little clearer.

Backblaze charges $5 per month for unlimited backup — so with operational costs of $100k per month (before employees), it needs just 20,000 users to break even. Backblaze didn’t provide us with its customer count, but its CEO did tell us that the company is profitable (Backblaze has 13 employees, incidentally, and it operates as a startup: i.e. everyone gets a tiny salary).

Another more prominent example is Dropbox, which charges $100 for 50GB of storage. Unlike Backblaze which runs its own hardware, Dropbox uses Amazon S3 for storage, which charges $0.055 per gigabyte per month for clients using more than 5 petabytes (Dropbox stores around 40PB). If the average Dropbox customer uses 25GB of his 50GB allotment, Dropbox pays Amazon $1.38 per month — and yet the service costs between $9 and $10 per month.

For cloud storage providers who operate their own data centers, such as Microsoft SkyDrive, Google Drive, and Amazon S3, the profit margins will be significantly fatter than Dropbox.

The caveat, though, is that all of these services have many more free users than paid. Dropbox, with 100 million users, is by far the largest cloud storage provider — but the vast majority of those are freeloaders. The number of paid users is probably quite small, and thus the profit margins are probably quite slim too.

Strategic dominance

The more likely reason for such feverish competition in the cloud space is strategic dominance; for Google and Microsoft, cloud computing is where the next platform war will be pitched. With Windows 8, you will log in with your Live ID and all of your documents and settings will automatically sync with the cloud, allowing you to freely roam between devices (or access your files from any web browser). Chrome OS is entirely web- and cloud-based, each successive version of Android develops deeper ties to the cloud — and of course, with Google+, the company is trying to weave as many of its web properties together as possible.

Ultimately, these companies are vying for control of your files. If you use Google Docs/Drive for word processing, then you’re much more likely to choose Chrome OS or Android, both of which are strongly tied into Google’s cloud — but if you use SkyDrive and Office Web Apps, Windows 8 and Windows Phone 7 make much more sense. In this regard, I’m not entirely sure what the future holds for “standalone” cloud storage services, such as Dropbox. If Microsoft, Google, and Apple can build cloud backup and file sharing into the operating system, what purpose does Dropbox serve?

Source

Google acquires Motorola Mobility, rolls the dice with Android

Some 9 months after Google’s announcement that it would rather like to make a subsidiary of Motorola Mobility, the $12.5-billion acquisition has finally been approved by the US, EU, and China. All eyes are now on Google to see what it will do with a bunch of software platforms under its belt — Android, Chrome OS, and Google TV — and its new-found top-notch hardware division.

Outwardly, we are promised, except for a new CEO — Dennis Woodside, former president of Google’s Americas operation, replaces long-time Motorola CEO Sanjay Jha — very little will change. The last 9 months have basically seen Google promising Android partners such as HTC and Samsung that Motorola Mobility will not get preferential treatment when it comes to future versions of Android or Nexus devices. The Chinese regulators even went as far as saying that they would only approve the acquisition if Google keeps Android free for other device makers for at least five years.

Beyond that, though, is for Google’s execs to know, and for us to find out. Officially, all we have to go on is a mysterious line in the acquisition press release that says Motorola will “enable Google to supercharge the Android ecosystem and will enhance competition in mobile computing.” In reality, there are two equally likely (but diametrically opposed) paths that Google could take. It is possible that Google simply wanted Motorola Mobility’s 24,500-strong mobile device patent portfolio to protect Android from further patent litigation. In this case, Motorola Mobility will continue to function exactly as before, but with new leadership. There are rumors floating around that Google will proceed to lay off a large number of Motorola employees, which would reinforce this speculation.

The other option is that Google will begin producing its own line of Android, Chrome OS, and Google TV devices. In this case, Google would effectively begin a transformation into a company that resembles Apple — but with Search under its belt, it might even have the edge on Apple. The main problem with this, though — putting aside Chrome OS and Google TV for the moment — is that Android devices are commodities. For Googola to compete with Samsung and other device makers, it would need to somehow differentiate itself. Selling “vanilla” devices won’t be enough, and I can’t see Google developing a custom skin, a la TouchWiz or Sense.

If Google does go the hardware route, I suspect it will fork Android — perhaps to create a commercial, non-free version that flawlessly integrates with Chrome OS, and other devices such as Android@Home and Google Glasses. Forking the codebase wouldn’t violate the Chinese stipulation that Android remains free for five years — but even so, is it really the best thing for the Android ecosystem as a whole?

For more, read David Cardinal’s take on why Google should sell Motorola as soon as possible.

Source

Google’s Knowledge Graph: Wikipedia on steroids, or the beginning of the end for the web?

Further establishing itself as the Borg of the web, Google is adding a Knowledge Graph capability to its search engine. Simply put, Google will now try and answer your question right on the search results page — no need to actually visit Wikipedia or your favorite travel information site when Google can simply absorb their data and present it to you nicely formatted along with a few choice ads. Google already has most of the video content on the web through YouTube, most of the geographic information through Maps and Earth, and an increasing share of the email, so why not? The user gets a quicker, and likely more useful, set of facts about their search.

At a glance, faster results are a no-brainer benefit for everyone. But like the spread of cheap, imported goods, there is a corrosive downside. Web publishers get cut out of the loop, risking their business models and ability to create the content that we all rely on. Google risks crushing the web in its embrace — unintentionally loving it to death. Looking more closely at Knowledge graph shows how the process works and why you may want to worry more than a little about it.

The Google Knowledge Graph — Baby steps to an information monopoly?

Google’s Knowledge Graph is brand new — “only” containing an estimated 3.5 billion facts about 500 million objects — but of course it will grow as rapidly as the Googleplex can organize additional information. Even now it is a powerful tool for those who want quick answers, and don’t like wasting their time surfing to get them — loosely described as semantic search. For comparison, Wikipedia currently has less than 30 million pages. When Google decides a search is about one of the 500 million objects it has categorized, it displays the facts it has about that object in a separate “knowledge panel” on the right of the page. Traditional search results and ads appear in the main body of the page on the left.

Google’s Knowledge Graph shows not just links for the Bronx Zoo, but a panel of facts on the right-hand side of the page

You can see here that Google has pulled out a map showing the location of the Zoo, along with some facts from Wikipedia — which to its credit are linked and attributed to Wikipedia. It also shows me other topics that those curious about the Bronx Zoo are interested in. Interestingly, the knowledge panel is less useful in this case than the first search result — which reports the top links from the Bronx Zoo site itself — but for many topics which don’t have a definitive website the knowledge panel is a handy place to get started with research.

Google vs. Bing: Do you trust the web or your friends?

While Google is using brute force to tame all the information on the planet in its effort to bootstrap the semantic web, Bing is taking a more surgical, and social, approach — by sifting through data about and from your friends to decide what might be important to you. Once you turn on Bing’s new Sidebar and sign into Facebook, Bing will happily crawl through any and all information it can find in your friends’ profiles or posts to cough up a variety of factoids. Some can be very useful — like a friend’s photo album from a place you’re interested in visiting, or the fact that someone has just reviewed a movie you’re curious about. Others are at best trivia — like the information that a long-lost business colleague used to live 100 miles from a place you’re considering for vacation, or that the musician “Taj Mahal” recently tweeted, shown in response to an attempt to find out which friends might know something about the Taj Mahal in India.

Comparing the two, so far I’d give the edge to Google and the Knowledge Graph. Useful snippets from friends (and I’ve got around 800 FB friends, so that should be a good sample) are few and far between in my efforts to use the new Bing sidebar. I can see that with time and improved linking technology the amount of useful information from friends will improve, but it’s hard to guess by how much. For my Bronx Zoo example, Bing’s sidebar coughed up a photo of a friend’s daughter, presumably taken at the zoo, as well as letting me know that two of my friends used to live in north New Jersey towns. None of it very useful in planning my event there.

Bing also offers an “ask friends” where I can ask my friends to help me with a search. Frankly, I’m not willing to even experiment with that. If I have a topic that I’m not sure how to approach, I’m old-fashioned enough to mail a couple friends who I think might be able to help out. So posting a search would only be useful if I could make it visible to just a few friends, instead of all 800 — but that doesn’t seem possible yet. Ironically, because of the way Google+ got started — with circles from the beginning — the same idea might actually work for me if Google implemented it so that I could ask a particular circle. Hopefully Bing will also allow the feature to be restricted to particular lists of friends.

One advantage of Bing’s Social Search over Google’s Search Plus Your World, at least for me, is that Bing clearly separates the social results from web search results. It is confusing and frankly a little weird to be doing a web search on Google and have various posts and articles written by me or my friends mixed willy-nilly into the results. Sometimes they’re useful, but other times they just get in the way since I’m really trying to look outward for new information, not navel-gazing by re-reading my old articles.

Knowledge Graph: The end of web publishing?

Like the snake that eats its own tail, there is a very serious problem with the way Google’s Knowledge Graph is likely to grow. Over time it will pull more and more information into its database — likely it has already swallowed the useful parts of Wikipedia — and give users less reason to actually traverse the web and visit any of the sites from which it has gotten its information. In turn, of course, that will starve those sites of needed revenue (or in the case of Wikipedia, attention and donations) and cause them to slow their acquisition and publication of knowledge. How long will it be before the “Report a Problem” feature of Google’s Knowledge Graph becomes more important to the web than submitting a correction through the arduous Wikipedia edit and review process? The resulting paradigm clearly isn’t stable.

This problem isn’t lost on Google, although its current answers aren’t particularly satisfying. Google’s executive in charge of search, Amit Singhal, says that as search engines improve, users perform more searches and also create more traffic to external websites. The trouble with that bromide is that in the past the improvements have been related to providing more accurate links to external sites — inviting increased browsing — and now they are being geared at providing answers directly on the Google site, which is an entirely different thing that might well decrease subsequent browsing.

Google’s head of Search, Amit Singhal, at SMX on the issue of how Google’s Knowledge Graph affects publishers.

Singhal also explains that to survive websites have to move further up the value chain, and not simply answer questions the search engine is able to. He uses the annoying and trivial example of a site providing the answer to “2+2.” Unfortunately, that answer shows the issue isn’t really deeply concerning to him, and apparently to Google. There is no “bright-line rule” beyond which Google won’t venture, only practical limits on its technology. This sounds very similar to the issue with PC utility software vendors providing services which are eventually bundled into the operating system. Realistically, it’s a warning shot across the bow of publishers that Google considers anything anyone wants to know as fair game, and if it can figure out how to provide that knowledge within its ecosystem and keep all the money — it will.

Once Google has effectively tied its Knowledge Graph into its digitized library of almost every book on the planet and scraped the contents of the semantic web into its Googleplex, it will have a practical monopoly on access to many kinds of information — even if you have a site with some other perspective, users will likely need to find it through Google. Public opinion, and in turn public policy, will get shaped by which factoids Google serves up in response to controversial searches like “climate change” or “intellectual property protection.”

Google Knowledge Graph results for global warming feature a well-known climate skeptic and an activist. Fair and balanced, maybe, but certainly a step towards editorial control coming from Google.

There is something more than a little insidious and even terrifying about this prospect. By declaring itself a repository of knowledge, rather than just an honest broker providing equal access to resources on the web, Google is making itself the sole arbiter of truth — or at least sole editor of the presented truth. Even Wikipedia provides an open, community-driven, process for editing, reviewing, and correcting facts as part of serving the community. If Google continues down its current path without articulating a clear set of transparent checks, balance, and access rules — or limiting itself to the role of a “common carrier” for the web’s information — it is likely we’ll soon hear a clamor for extending anti-trust regulations to limit monopolies on the access to knowledge.

Read more about Knowledge Graph, or the semantic web

Source

Why Diablo 3′s always-online DRM is a good thing

Diablo 3 has now been available worldwide for a week. In true, zealous, Blizzard-super-fan style, the end boss on the hardest difficulty setting has already been defeated. We don’t know the exact number of Diablo 3 sales, but with some 2 million pre-orders, we’re talking about one of the fastest-selling games of all time — and probably the fastest-selling PC game of all time (beating out the World of Warcraft expansions and Starcraft 2, which are also Blizzard titles).

By almost all accounts, Diablo 3 is a fantastic game. It has received almost perfect scores from the top game critics, hardcore gamers, and casual players alike — except for one niggling issue: Diablo 3 requires an internet connection to Battle.net to play.

It’s not like this is news — Blizzard announced that the game would require an internet connection last year — but unfortunately it was brought into sharp relief by a flurry of server issues and extended maintenance downtime over the past week. If there hadn’t been any login difficulties, the always-online thing probably would’ve faded into the background, obscured by legendary-tinted glasses — but because some 2+ million people have spent the past week battling with heinous lag, Error 33, Error 37, and Error 3003, critics and gamers have focused their ire on what they perceive to be the issue: Diablo 3′s always-online DRM.

Now, while it’s true that D3′s reliance on Battle.net does act as a form of DRM, this is just a value-added extra as far as Activision-Blizzard are concerned. Really, given the faith that Blizzard has garnered from its users over the last 20 years, I’m surprised that fans have been so quick to label this feature as crippling DRM, or worse, the hellspawn of Blactivision’s CEO, Bobby Kotick. In reality, Diablo 3′s connectivity requirement exists for a handful of compelling reasons that actually make the game better.

Gameplay

Users and critics are quick to draw comparisons to Ubisoft’s aborted use of must-remain-connected-to-the-internet DRM, but Diablo 3 is really very different. To begin with, it helps if you think of Diablo 3 as an online game. In fact, think of Diablo 3 as an MMO, like World of Warcraft or Star Wars: The Old Republic.

Unlike Diablo 2 or other offline games, all of your D3 characters reside in the cloud — in one of Blizzard’s huge database servers, just like your WoW character. Every time you loot or equip an item, that change is automatically saved to Blizzard’s database. Because of this, your characters will survive a hard disk crash, and they’re accessible from any internet-connected computer — you can log into D3 from a friend’s computer (or your office computer) and continue from where you left off.

Once you actually log in, Diablo’s gameplay is a piece of seamless mastery. Joining a friend’s game takes one click — and then porting directly to them takes one more. All your characters share a single item stash. With the BattleTag and Real ID systems, you can set up a new game with two or three clicks — and of course, like SC2 and WoW, you can chat to any of your friends, no matter which game they’re currently playing, from the chat window.

None of this would be possible without permanent Battle.net connectivity.

Security

One of the biggest problems with offline games is that all of the game logic and game saves must be stored locally. Game saves are historically very, very easy to crack — outfitting your character with a full set of end-game legendary items can be as simple as loading up a hex editor and changing a few values. With decompilation and memory interrogation hacking, a game’s inner workings — boss mechanics, loot generation, and so on — can be laid bare.

In offline-only games, these factors aren’t a big issue — but in Diablo 2, which featured offline and online play, so much data was available in the game client that hacked characters and map hacks were rife.

Diablo on the left, WoW on the right — the look similar because they are

To put it another way: Can you ever imagine an MMO like WoW storing character data locally on your hard drive? No. Just apply the same logic to Diablo 3. Blizzard wants to keep hacking and exploitation to a minimum — and hacked games and characters were one of the biggest complaints in Diablo 2 — and so it isn’t all that surprising that the company is keeping as much data and logic server-side as possible.

The real-money auction house

The jewel in Diablo’s crown — and one of the most exciting experiments in online/MMO history — is the real-money auction house (RMAH). When it launches at the end of May, players will be able to sell items for actual dollars and pounds, with Blizzard taking a flat $1 cut on unique items, and 15% on stackable commodities (health pots, gems).

The real-money auction house is Blizzard’s attempt to monetize (and legalize) the rampant, gray/black market buying of gold and items from “Chinese gold farmers” using real money. Similar approaches at stemming gray/black market trading have been tried before — most notably, Eve allows you to buy your monthly subscription with in-game currency — but Diablo 3′s RMAH is by far the highest-profile attempt yet. The intricacies of virtual world economies are beyond the scope of this story, but put it this way: It is actually more efficient for someone in China (or another developing country) to farm gold in WoW and sell it to rich Westerners, than to work in manufacture or agriculture.

Diablo 3 real-money auction house

It is likely, if Diablo 3 attracts enough players, that a huge real-money economy will develop around the game, making Blizzard (and players!) millions of dollars in the process. For a game without a monthly subscription, this is obviously a rather attractive proposition for Blactivision.

For the RMAH to work, though, Blizzard needs absolute security. Dupe bugs and hacks — where players can magically conjure up duplicates of the same item — could ruin the legitimacy of the RMAH overnight. This isn’t to say that MMOs are perfectly secure (almost every big MMO has fallen victim to a dupe bug), but they’re a lot more secure than offline games. As we mentioned previously, every action that you perform in Diablo 3 is recorded and saved in one of Blizzard’s databases. It might be possible to dupe some gold or items, but Blizzard will have built in plenty of protections that allow it to track the duped objects and remove them from the game. There is almost certainly a team at Blizzard that is tasked with hunting for exploits and exploiters 24/7.

Epilogue

In my opinion, these three factors are a large part of what make Diablo 3 so damn enjoyable. With that said, though, there is simply no avoiding the fact that requiring an internet connection to play is ultimately very lame. DSL connections are finicky beasts, after all — and what happens if Blizzard gets DDoSed, or it goes bankrupt? As internet connections improve, though, these concerns will slowly vanish.

In many ways, Diablo 3 is almost the first step towards cloud gaming. While I can’t confirm this, I suspect that the D3 artwork is the only real asset that is stored on your computer — everything else is either downloaded, or streamed while you play. You will notice that, despite the tumultuous first week, Blizzard hasn’t issued a single patch for D3. It’s entirely possible that, when you log in, item stats, monster attributes, spell effects, and even game mechanics are downloaded from the Battle.net servers. It also seems that Blizzard has the capability to change items, monsters, and spells while you play the game. In effect, you are streaming everything from the cloud — except the art.

I think this is the real reason that Diablo 3 completely forbids offline play. Blizzard could make it so that you could download a complete version of Diablo 3 for offline play, but a) that would give hackers easier access to data that Blizz would rather keep locked up; b) the ability to continually fine-tune the game after launch is very desirable; and, finally, who are we kidding: As far as the accounting bods are concerned, online-only play is the single best way of cutting down on piracy and generating more income — at least until someone reverse engineers the Diablo 3 server, of course.

Read more about Diablo 3, or the awful DRM in Batman: Arkham City

Source

Can probabilistic computing save Moore’s law?

At the ACM International Conference last week, researchers from Rice University, Singapore’s Nanyang Technological University, Switzerland’s Center for Electronics and Microtechnology, and UC Berkeley unveiled a microprocessor that’s designed to get the answer wrong — at least, some of the time. The idea of building microprocessors that deliberately allow for incorrect results has been kicking around for years; the prototype silicon demonstrated last week is the first time anyone has demonstrated the idea in native hardware.

The problems facing conventional CMOS scaling are well-documented — could PCMOS (Probabilistic complementary metal-oxide-semiconductor) provide an answer? As the name implies, it’s an approach that’s compatible with conventional CMOS manufacturing. The Rice paper claims a 15x power improvement over normal silicon — let’s take a look at why.

Nearly all of the problems facing conventional semiconductor scaling are rooted in the need to control variance, or noise. Noise, in this context, is defined as “unwanted additions to a signal.” It’s impossible to eradicate and literally universal; we measure the apparent age of the universe by measuring the distribution of cosmic background radiation. In semiconductor manufacturing, unwanted variance has become a major problem in multiple ways. Electrical noise forms a lower boundary to CPU voltage — below a certain threshold, the “On” signal that tells a transistor to flip from a 0 to a 1 becomes lost in the background. Thus, lower voltages can only be achieved in concert with lower noise levels — and achieving those levels becomes exponentially more difficult as process nodes shrink.

PCMOS flips (PDF) conventional thinking on its head. Instead of trying to generate and measure a distinct signal that stands out clearly from noise, why not measure the noise distribute and, from that, detect the presence or absence of a signal? Such an approach can only be effective if the system is designed to be capable of tolerating incorrect conclusions, but the power savings in such cases can be significant.

We borrowed from one of the simpler graphs to show the potential energy savings of a probabilistic design. The y-axis shows the amount of power required to flip a probabilistic switch, the value pdenotes the chance that the switch flips accurately. The legend reflects base voltage and the type of workload being run.

At the far end, the graph demonstrates the “cost” of perfection. Ensuring that p always flips properly requires an input of ~1.2E^10-13, while flipping p at a 90% accuracy rate requires nearly a full order of magnitude less power. This is legitimately cool stuff, and we’re going to talk more about deployment and usage — but it’s not the fundamental breakthrough or Moore’s law savior it’s been portrayed as in some circles.

Perfection has always been expensive

Back in the early days of computing, if you wanted a dedicated hardware unit for handling floating-point math (math with a decimal point), you had to buy a separate chip that fit into a socket on the motherboard. Intel designed an FPU for both the original 8086 and the 286 that followed; these were dubbed the 8087 and 80287, respectively.

When it built the 80387, Intel elected to do something unusual — even though it took the company until 1987 (two years after the 386 launched) to bring the FPU to market. The 80387 FPU was the first x86 chip to be fully compliant with the IEEE-754 standard for floating point arithmetic. It’s a useful example in this case, because Intel was willing to go to significant trouble to design a chip that would meet the standard at a time when die space was at a much higher premium than it is today.

The x87 FPU provided (and still provides) single-precision (32-bit), double-precision (64-bit) and extended precision (80-bit) operating modes. What may surprise you is that by default, the x87 FPU used all 80 bits in order to guarantee sustained precision over many operations. One of the chief designers of the IEEE-754 standard, William Kahan, noted that the standard was designed to “to serve the widest possible market… . Error-analysis tells us how to design floating-point arithmetic, like IEEE Standard 754, moderately tolerant of well-meaning ignorance among programmers.”

One of the reasons CPU designers have pursued tighter tolerances and greater robustness is that it’s almost always cheaper to get the answer right the first time than it is to have to re-do the problem starting from scratch. Even simple, single-bit error correction adds a degree of complexity; true validation logic that performs independent analysis and verification of results generated elsewhere is extremely expensive and difficult to integrate.

The work being done at Rice won’t change that. It’s not going to re-write Moore’s law or fundamentally change the shape of CPU design. That doesn’t mean it’s not important.

When we don’t need to get it (all) right

The chief market for these types of chips is in areas where the cost of being wrong or of having to perform a calculation again is low, while the power savings is potentially quite high. One obvious application is in video signal processing, particularly if the characteristics of the architecture can be fine-tuned to make mistakes in places where the human eye is biologically less-equipped to notice them.

Here’s an example of how the Rice University prototype renders video compared to a conventional image (far left). The middle image is rendered with an allowed error rate of 0.54%; the far-right is rendered with an error rate of 7.54%. Rice University claims that the far right hand image required 1/15 the power of the far left.

There are substantial caveats to this. We don’t know what hardware was used to create the baseline image and were unable to secure a copy of the paper or speak to the researchers in question. Clearly software and hardware would need to be fine-tuned; it’s entirely possible that more robust software correction might allow for an improved picture without dramatically increasing power consumption.

Rice’s work is a first step, not a final iteration. It can’t replace conventional CPUs; even the I-Slate’s being built in Singapore for Indian markets are identified as “containing” a probabilistic chip, not unilaterally relying on one for all system functions. The available papers and theses all describe a PCMOS processor as a specialized accelerator. Intel thinks the idea has merit; the company has previously funded research into the PCMOS concept.

Intel isn’t publicly talking about PCMOS, but the company has its own plans for reducing power consumption by lowering the number of bits used in certain kinds of operations. The approach is different, but the idea that maybe we don’t need to be quite as accurate, in every scenario, is coming under increasing scrutiny. PCMOS uses it as a means to reduce power consumption at the transistor level; Intel is exploring using fewer bits in certain cases, but they’re both specialized approaches to bringing power consumption down — and improving battery life in a future generation of products. PCMOS won’t reinvent the wheel, but it could provide much-needed axle grease in future product generations.

Source