Tech Talk

The competition for cooler servers is heating up. This week, Austin-Tx.-based start-up Calxeda announced it received another US $55 million in funding, which could help propel its chips into a server market that's struggling to keep down power consumption. But the company faces steep competition.

Founded in 2008, Calxeda made waves last year when Hewlett-Packard announced it was working with the company to develop a new line of low-power HP servers. Calxeda’s server chip is based on 32-bit designs from ARM, which licenses IP to nearly every smartphone chipmaker. The company reckons servers built with its chips will consume much less power and space than those built with today’s 100-Watt behemoths. 

This latest cash infusion, which includes investments from Austin Ventures and Microsoft co-founder Paul Allen’s Vulcan Capital, is an added vote of confidence for Calxeda and a sign that the company might just be on to something.

When I spoke with Karl Freund, Calxeda’s VP of Marketing earlier this year, he said low-power chips will help offer more choice in a server industry dominated by high-speed options: “What you’ll see is the industry will go from a one-size-fits-all model to tailored solutions, much in the same way you see in the cell phone business.”

Freund said Calxeda isn’t aiming for the entire server market. Instead the company is looking at a growing segment of server applications, like processing web search results, that can be broken down into bite-sized computational chunks. Those sorts of workloads can be “scaled out,” meeting increased demand by simply adding more processors.

But Calxeda isn’t the only company with this idea. It's not even the only company whose chips Hewlett-Packard is using for this idea. HP is moving forward with similar “microservers” made with low-power Intel Atom chips. Then there is Sunnyvale, Calif.-based chipmaker AppliedMicro, which is pursuing 64-bit server chips, also based on ARM designs. Even Samsung seems to be getting into the game with its own ARM-based CPUs. It will take a while to see how these bids all shake out, but the next couple of years are shaping up to be pretty interesting.

(Image: Calxeda)

An abundance of cheap, renewable energy, particularly hydropower and geothermal, has drawn aluminum smelters to Iceland. It's become an industry that already consumes five times as much electricity as the country’s residents, and more aluminum plants are on the drawing board—raising concerns about how much the country’s economy is relying on one industry.

Meanwhile, there is another fast-growing, power-hungry industry in the world: cloud computing and storage. “The cloud” seems so light and fluffy, but building a cloud involves huge clunky buildings full of servers. Just one of these server farms, according to an April report by Greenpeace, can consume the energy equivalent of 180 000 homes. The companies that run them do their best to be efficient, because high energy costs hurt profits—and also, in some cases at least, because of a corporate commitment to the environment. The April Greenpeace report praised Yahoo and Google for “prioritizing access to renewable energy in their cloud expansion” but criticized Amazon, Apple, and Microsoft for rapidly expanding their clouds “without adequate regard to source of electricity,” relying “heavily on dirty energy.”

Which brings us back to Iceland. Even with all that aluminum smelting, Iceland has a renewable energy surplus. And, since the recent addition of two new, high-speed, transatlantic fiber optic cables to the country’s single older fiber cable (and one more going into service soon) it’s got bandwidth to spare as well.

It turns out Iceland also has entrepreneurs with big ideas ready to take advantage of this power and bandwidth, such as GreenQloud, which says it's ready to offer commercial cloud services. At the DemoFall conference last week in Santa Clara, Calif., cofounder Eirikur Hrafnsson described how he started working on the idea that became GreenQloud in 2008, after seeing a Gartner report indicating that the IT industry is responsible for as much greenhouse gas generation as the aviation industry—some 2 percent of the world’s carbon emissions. A McKinsey report around the same time predicted that this number would double by 2020.

Hrafnsson says his company wants to take on Amazon, currently the go-to company for businesses and government entities that want to offload their computing to the cloud. Will big organizations really trust their data to a little startup in Iceland? Hrafnsson is betting that a trifecta of attributes—greener, cheaper, and what he says is a better software platform—will inspire  potential customers to give GreenQloud a chance.

Photo: Gulfoss, Iceland. Source: Wikimedia Commons

Follow me on Twitter @TeklaPerry.

David J. Wineland of the United States and Serge Haroche of France will be awarded the Nobel Prize in Physics  “for ground-breaking experimental methods that enable measuring and manipulation of individual quantum systems”.

Affiliated with the National Institute of Standards and Technology, Wineland’s work allows for “optical” clocks that reach a precision of 10^-17 seconds, a hundred times better than the caesium clocks that set the official time in the United States. Optical clocks use a self-referential technique—one ion is used as a clock, another is used to read the clock without altering its fragile quantum state. Their precision allows for measurement of the effects of relativity—like time dilation and gravitational shifts—even across distances as small as tens of meters. So far optical clocks have run in Wineland’s lab for “many hours and days,” he said in an interview on the Nobel Prize’s web site.

Wineland’s group has also demonstrated computing operations based on two quantum bits. Unlike conventional computers, where the basic unit of information, a bit, can take the binary values of either 1 or 0, a quantum bit can be 1, 0, or both 1 and 0 at once. Computation scales up dramatically, because n quantum bits can represent 2ⁿ states at the same time, theoretically allowing for unimaginable computational power. “Most of us feel that even though that is a long, you know, long way off before we can realize such a computer,” Wineland said, “many of us feel it will eventually happen.”

Both Nobel Laureates’ work probes the quantum properties of particles in isolation. In particular they explored a quantum phenomenon called superposition, where a particle can be in two states at once. The phenomenon was made famous by Austrian physicist Erwin Schrodinger’s thought experiment. A cat—for some reason usually shown as black—is isolated in a sealed box that also contains a vial of poison. The poison is released when the decay of a radioactive atom in the box is detected. Because radioactive decay is a quantum mechanical process, there is a level of uncertainty in when the atom decays. The system is in a state of superposition—the atom has both decayed and not decayed, so the cat is both alive and dead. Opening the box “collapses” the state and possibly kills the cat.

Schrodinger didn’t think that it would be possible to study this collapse of a quantum state in detail. In 1952, he wrote, “We never deal with just one electron or atom, or (small) molecule,” except in thought experiments.

But the work done by the Laureates did just that, describing the “progressive collapse” of the wave function of a single particle. Haroche and others have even created “cat states”.

Wineland and Haroche’s techniques are neatly complementary: While Wineland traps electrically charged atoms using laser light, Haroche measures trapped photons by sending atoms through a trap.

Haroche, who noticed the 46 Sweden code when the Nobel call came this morning, said in an interview on the Nobel Prize website that the ability to work with single atoms and photons means that quantum properties that are “veiled” due to statistical effects, come out in the open. “If you were to ask me what was the application,” he said, “I would tell you I don't know. And I would just tell you that I think there will be some applications.”

This story was corrected on 10 October.

In August 1479, a Franco-Portuguese fleet attacked a Genoese merchant convoy just off Cape St. Vincent, the southwest tip of the Iberian Peninsula. Three Genoese ships went down, leaving one wounded, almost-twenty-five-year-old sailor-adventurer adrift clutching an oar. He swam ashore at Lagos, Portugal, to find himself in the pilot house of the Age of Navigation.

And so the Genoan castaway Christopher Columbus started to soak up the high technology of his time, learning math, celestial navigation, shipbuilding, Latin…and map-making, the great enabling technology of the epoch. In short order, Columbus and his brother, Bartholomew, started (probably) a chart-making business in Lisbon, an enterprise that put him in touch with information and ideas from all of Europe. It was an information age, and the information was generating huge discoveries and massive fortunes—like those produced by the high tech and biotechnology booms of our own time.

So today we’re dipping back into Samuel Eliot Morison’s wonderful Admiral of the Ocean Sea (with the combinatorial publication date of 1942) for a look at how measurement—or, more particularly, mismeasurement—fueled the explorer’s conviction that he could reach the Indies by sailing west across the Atlantic. His certainty, rooted in some of his age’s best measurements, technologies, and calculations, was undermined by the entrepreneur’s hallmark character trait: When faced with several possible values for a key variable, Columbus would invariably choose the most optimistic.

We thus owe the great 1492 Enterprise of the Indies to three serious measurement errors.

Washington Irving’s overly imaginative A History of the Life and Voyages of Christopher Columbus notwithstanding, it was widely known by the 15^th Century that the Earth is spherical. The question was, how big is the sphere? In 200 BCE, after all, Eratosthenes calculated the circumference of the earth to within one percent of its actual girth. He figured that one degree of latitude was equal to 59.5 nautical miles.

In making his own calculation, however, Columbus preferred the values given by the medieval Persian geographer, Abu al Abbas Ahmad ibn Muhammad ibn Kathir al-Farghani (a.k.a. Alfraganus): one degree (at the equator) is equal to 56.67 miles. That was Columbus’s first error, which he compounded with a second: he assumed that the Persian was using the 4 856-foot Roman mile; in fact, Alfraganus meant the 7 091-foot Arabic mile. (This is, of course, the sort of confusion of units that sent the Mars Climate Orbiter into its terminal swan dive in September 1999.)

Taken together, the two miscalculations effectively reduced the planetary waistline to 16,305 nautical miles, down from the actual 21,600 or so, an error of 25 percent.

And then there was the third error. “Not content with whittling down the degree by 25 percent,” Morison writes, “Columbus stretched out Asia eastward until Japan almost kissed the Azores.” Through a complicated chain of reasoning that mixed Ptolemy, Marinus of Tyre, and Marco Polo with some “corrections” of his own, Columbus calculated that he would find Japan at 85º west longitude (rather than 140° east)—moving it more than 8,000 miles closer to Cape St. Vincent.

All in all, he figured, the Indies were just 68 degrees west of the Canary Islands. Calculated travel distance: 3080 nautical miles. Actual distance from Tenerife to Jakarta: 7313 nautical miles. Margin of error: 58 percent.

“Of course,” Morison noted, Columbus's “calculation is not logical, but Columbus’s mind was not logical. He knew he could make it, and the figures had to fit.” Morison, an admiral himself, is full of admiration for Columbus’s skill as a practical navigator, capable of pinpoint landfalls on his returns to the New World. As a metrologist and theoretician, however, he failed to double-check his work.

Image: Douglas McCormick

According to the U.S. National Highway Traffic Safety Administration, there were 5.4 million automobile crashes on U.S. roads in 2010, killing 33 000 people and injuring more than 2.2 million. In a paper recently published in IEEE Transactions on Intelligent Transportation Systems, two researchers at Virginia Tech’s Center for Injury Biomechanics delve into just how much of an effect systems that warn a driver about an impending front collision—then slam on the brakes if the driver doesn’t act quickly enough—might have on these crash statistics.

Automakers are starting to introduce vehicles equipped with electronic safety systems whose purpose is to keep cars from crashing. The researchers, Clay Gabler, a professor of biomedical engineering, and Ph.D. student Kristofer Kusano, studied a suite of systems that rely on radar to tell the car when it is coming dangerously close to another vehicle’s rear bumper. Some of these systems deliver an audible warning when the distance between the car and the one ahead of it gets too narrow.

Others offer braking assistance if the driver responds to the warning by applying the brakes. Still another type attempts to bring the car to a halt with a huge braking force if the driver has not hit the brake pedal 0.45 seconds before the sensors predict that there will be contact.

Gabler and Kusano combed through 5000 investigator reports of crashes. In computer simulations that recreated the scenarios of 1400 rear-end collisions (for which investigators from the U.S. Department of Transportation had gathered information such as photographs and diagrams of the crash scenes, police, driver, and occupant statements, and vehicle damage assessments), the Virginia Tech researchers were able to demonstrate the extent to which the electronics would have helped. They concluded that in most cases, the electronic safety systems would slow cars down enough to cut the number of serious injuries in half. Better still, they say, 7.7 percent of rear-end collisions would be avoided altogether.

“Even if the driver is distracted and does nothing, a system of this type would brake forcefully enough during that final half second before impact to slow a car traveling at [72 kilometers per hour] by about [10 to 12 km/h],” says Clay Gabler, who is also assistant director of the Center for Injury Biomechanics. “That might not seem like a lot,” he says, “but the aim is to reduce the energy of a collision. And since kinetic energy is related to the square of velocity, this change in speed reduces the likelihood of serious injury by about 35 percent. That’s huge.”

Bedeviled by the challenges of measuring nanofeatures on ever-shrinking chips? A new hybrid statistical approach might let you combine a variety of optical methods to improve measurements, reduce uncertainty, and boost throughput.

A team at the National Institute of Standards and Technology had developed what they call a rigorous approach to combining different metrological technologies—scatterometry (calculations of physical dimensions from scattering patterns of electromagnetic waves) and atomic force microscopy (AFM), for example—to produce measurements that are more accurate than either method can deliver individually. (Researchers Nien Fan Zhang, Richard M. Silver, Hui Zhou, and Bryan M. Barnes reported their results in Applied Optics.)

They tested their approach on optical critical dimension (OCD) measurements of arrays of etched ridges on silicon chips 60-70 nanometers tall and about as wide at the base. (Sixty nanometers is very roughly 300 silicon-atom diameters.)

At this scale, there is no ruler. Scatterometrists generate curves—typically, of the intensity of reflected light—for a wide range of input variables (wavelength, incident angle, or polarization, for example). They compare these data to a much wider array of curves derived from models—each having a particular combination of, for example, line height, width, edge roughness, sidewall height, sidewall angle, etc. When they find a best-fit match between the real-world curve and the theoretical curve, they figure that the real-world array pretty much matches the model. The measurement isn’t exact, of course—there are always uncertainties and noise in the system, and different parameter combinations can yield similar results.

In testing their approach, the NIST researchers started with scatterometric measurements. In one case, for example, they made four scans of a silicon nitride line array etched on a chip. Each scan plotted light intensity for 21 incident angles, changing the polarization or the orientation of the light source with each scan.

The best fit to the parametric model indicated that the top of the trapezoid measured 33.7 nm wide (with a 10.8 nm standard deviation), the middle width (halfway between base and top) was 48.9 nm (6.0 nm), and the height was 60.0 nm (2.2 nm).

A subsequent AFM measurement gave somewhat different readings: top, 37.6 nm (0.9 nm); middle, 48.0 nm (1.9 nm); height, 57.5 nm (0.7 nm).

Enter Thomas Bayes (with an assist from Pierre-Simone LaPlace) and Bayes’ Theorem—the mathematical tool for refining initial assessments of the probability of an event (for example, that a measurement will produce a certain value) with information from later or other events.

The approach seems intuitively obvious, at first—experience is always reshaping our assumptions after all. And, indeed, iterative approximations are no strangers in the math world. But there’s something about real-world applications of Bayes’ Theorem that seems, well, spooky—like something derived in the arithmancy department at Hogwarts. I’m focusing on Event A, right here in front of me, but my expectation of results can be changed greatly by a null result for Event B, over there on the other side of the room. It can seem as maddening as the Monty Hall  problem...but more complicated. (For an entertaining, if sometimes unmathematical, history of Bayes’ Theorem, see Sharon McGrayne's book, The Theory That Would Not Die.) 

So the researchers (under the mathematical guidance of statistician Nien Fan Zhang) applied a Bayesian manipulation to the scatter data, using all of the AFM measurements to modify each of the scatterometric data points. The result: top width, 38.0 nm (0.9 nm); middle width, 48.9 nm (1.8 nm); height, 58.6 nm (0.4 nm).

There are three take-aways. The combined Bayesian measurement has a higher probability of being correct than either the scatter or AFM value alone. The expected error is much smaller (improving on the scatterometric uncertainty by more than an order of magnitude in once instance). Finally, one would expect that combining measurement methods (especially a relatively demanding method like atomic force microscopy) would slow down the measurement process. Not necessarily, the authors say: Well-chosen sampling strategies can, in fact, improve measurement throughput—all while improving measurement and reducing uncertainty.

The approach, the paper concludes, is suitable for a wide range of new metrological combinations—combining model-based scanning electron microscopy, quantitative ellipsometry, and other methods. Overall, “the new hybrid methodology has important implications in devising measurement strategies that take advantage of the best measurement attributes of each individual technique.” And, indeed, researcher Richard Silver says that at least two manufacturers have deployed the hybrid methodology and are already improving their measurement results.

This past Thursday, I attended the third annual Open Hardware Summit, organized by the Open Source Hardware Association and held at the Eyebeam Art + Technology Center in Manhattan. While open software is now very much mainstream, open hardware is in a far more primitive state. So hearing from the folks at ground zero of this newfangled way of developing and marketing products was illuminating.

Before recounting some of the highlights of this conference, I should take a moment to try to outline what open-source hardware is. The basic concept is simple enough: It’s hardware for which the design documents—schematic diagrams, board layouts, CAD files, whatever—are all made available to anyone under some sort of open license. As with open software, different types of licenses grant varying degrees of freedom (although a lot of freedom appears to be the norm here). This approach stands in stark contrast to the usual way of doing business, where a company encircles its intellectual property wagons and keeps competitors at bay with a variety of weapons: copyrights, patents, or simply by maintaining trade secrets.

To anyone over about 21, it’s difficult to get your head around how a business can possibly make a buck when it gives most of its intellectual property away for bupkis. But some really do. Sparkfun is probably the poster child for this movement, having grown over the past decade from a dorm-room operation to a multi-million dollar business.

The solution to the paradox seems to be in the implicit assumption—that giving something away for free provides nothing in exchange. In fact, such generosity can bring a company quite a lot. For one, it fosters customer loyalty. And these loyal customers don’t just offer repeat sales. Many offer advice about the new products or product modifications they want. Some of the more talented offer their services to improve existing products and designing new ones, providing a break-neck pace of innovation. And more than a few offer to help other customers out. So a company gets quite a bit for “free,” reducing the usual expenditures it would have to make in product development and customer support.

Enough of the explanatory preamble. What did some of the people who really know about this subject have to say at Thursday's summit?

Keynote speaker Chris Anderson kicked off talks with remarks about open-source hardware in general and his open-source hardware brainchild DIY Drones in particular. Anderson, for those not familiar with his energetic schedule, is a prolific author and the editor-in-chief of Wired magazine; he also founded and helps run DIY Drones, a virtual get-together for tinkerers who build their own unmanned aerial vehicles for fun. Well, it's that plus a healthy online business that sells open-source autopilots and other kinds of hardware to support this high-tech pastime.

Anderson spoke about some of the things he and his colleagues at DIY Drones have had to grapple with—just how open to be with everything, how to reward those people in their community who make the greatest contributions back to the company, and how to respond to “cloners” who take the designs developed at DIY Drones and sell identical hardware at cut rates.

Such copying is, of course, allowed--although cloners, Anderson explained, often go over the line by expropriating copyrighted marketing materials as well. This might seem the Achilles heel to any open-hardware-based business. But, Anderson argued, cloners can’t duplicate an open-hardware business’s chief asset: the community it builds around it. He also told a charming story of how he brought one Chinese cloner into the DIY Drones fold, where the would-be competitor began to contribute—first translations, then bug fixes, eventually becoming a DIY Drones lead developer.

Another highlight of the day was a talk by “Akiba” of Freaklabs (Cher- and Madonna-like, he goes by just one name), who recounted how he and a few buddies from the Tokyo Hackerspace made good use of open-source hardware in the aftermath of the Fukishima reactor disaster. At a time when Japanese government officials were assuring people outside the evacuation zones that radiation levels were safe, Akiba and friends were cobbling together radiation monitors to double check those assurances. In some areas, they found the government’s rosy picture to be grossly misleading. Borrowing from various open-source designs, they quickly assembled a half-dozen or so geo-tagging, data-logging radiation monitors, which they strapped onto their cars and began crisscrossing the nation, ultimately uploading millions of readings to the Safecast database.

While the summit included many other speakers and many other entertaining stories of their open-hardware exploits, I’ll leave this report with the As—Anderson and Akiba. Their presentations bracketed the spectrum of applications for open-sourcing hardware—from building a for-profit business to altruistically trying to help out some fellow human beings in peril. The day’s especially refreshing take-away, though, was that, with open sourcing, the line between such very different pursuits can be a lot fuzzier than we’re used to.

So, in addition to scales that can weigh single protein molecules to an accuracy of a few hundred times the mass of a proton, we now have a position sensor that can detect changes in displacement as small as 1/30,000^th of the diameter of a single carbon atom, or about the diameter of a single carbon nucleus.

Houxun Miao, Kartik Srinivasan, and Vladimir Aksyuk (researchers at the National Institute of Standards and Technology’s Center for Nanoscale Science and Technology) have built a microelectromechanical system (MEMS) that pushes the lower limits of distance and force measurement down to within a short hail of the theoretical limits—to 2.3 times the standard quantum limit.

The device is a silicon-on-oxide chip fabricated via electron-beam lithography and reactive ion etching to leave a 15-micrometer-diameter silicon disk, a waveguide and actuators standing in relief from the chip surface (see scanning electron micrograph). Atop this structure, the researchers vapor-deposited a layer of silicon nitride that is patterned and etched to produce a 19-micrometer-diameter SiN_x ring suspended at the end of the actuator arm above the silicon disk. The actuator arm is a springy cantilever beam; force on the nitride ring pushes it down towards the disk. Researchers can adjust its vertical position by changing the potential between the beam and the chip base. This lets them both calibrate the beam’s spring constant and fine-tune the distance between the nitride ring and the disk. With this information, they can detect displacement with a precision of 4.6 femtometers—4.6 x 10^-15 meters—and a forces to within 100 attonewtons (about 10^-16 Newtons).

Both silicon and silicon nitride are transparent to infrared light. This means that the Si disk can act as an optical cavity—a photonic hall of mirrors that reflects light almost endlessly around its perimeter. “The light can be trapped inside the disk, close to its outer edge, and travel around in circles inside,” Aksyuk says. “It is trapped the same way at the light is trapped inside optical fibers used for transmitting light over long distances, but here it just goes in circles.” It is, he says, similar to an acoustic whispering gallery, which traps sound along a concave wall.

The key is the “evanescent tail.” Even when light is fully trapped in a channel—a fiber optic cable, a waveguide, or a silicon disk--its wave function extends outside beyond the channel’s borders. That’s the evanescent tail. The NIST device positions three components—the waveguide, the silicon ring, and the silicon nitride disk—close enough together so that their evanescent tails can interact.

The researchers tune light so that its wavelength is an integral fraction of the disk’s circumference. When they pump it from the waveguide (the pipe running from lower left to upper right of the SEM) into the disk, it will resonate in the optical cavity. “Under this condition the optical power inside the disk is increased dramatically—by up to 370,000 times in our case,” Aksyuk says.  

Any displacement of the nitride ring changes its interaction with the disk’s evanescent tail (the blue dots at the rim of the disk in the schematic), diverting some of its energy into the ring and reducing the resonance frequency. The researchers read the new resonant mode, and the frequency change indicates the change in displacement.

It’s important to note that, though future versions will likely be cryogenically cooled, the test device operated at room temperature, where thermal noise is a major problem. The NIST group coupled a “cold-damping” loop into the actuator controls, feeding the ring’s noise oscillations back with a phase delay to counteract the jitters. Think of it as the world’s smallest noise-cancelling headphone.

Images: V. Aksyuk, NIST

Like many start-up companies, business networking site LinkedIn once struggled to capitalize on the mountains of data generated by its users. Then, in 2006, a new hire, data scientist Jonathan Goldman, swept in and tamed the unwieldy data mess in a way that launched the company to the next level. Goldman extracted patterns from the connections between LinkedIn's users, and came up with a way to suggest to those users other people they may know. The "people you may know" feature created millions of new page views, and LinkedIn's growth went skyward. 

Sexy? The folks at Harvard Business Review think so, and in their October issue they proclaimed the data scientist the sexiest job of the 21st century. The authors of the article, Thomas H. Davenport, a visiting professor at Harvard Business School, and D.J. Patil, a data scientist at Greylock Partners, liken the profession to the Wall Street quants of the 1980s, and the computer engineers of the 1990s. "If 'sexy' means having rare qualities that are much in demand, data scientists are already there," they wrote.

A data scientist's job description goes like this: Make discoveries while swimming in data. Possess an intense curiosity. Bring structure to formless data and make analysis possible, all while having a feel for business issues and an empathy for customers. Advise executives on how to use the information to make better products.

What kind of professional can do all of these things? The rare kind with the powerful combination of skills that let them wear the hats of data hacker, analyst, communicator, and trusted adviser—all of which must be applied to a specific technology or product. (Spectrum last year identified 26 variations of data mining in IT).

Because more and more companies need someone with these skills, the demand for data scientists is exceeding the supply. These professionals garner high salaries and large stock option packages, the authors found in an informal survey. But more than that, data scientists want to be "on the bridge"-—a reference to Star Trek, in which Captain James Kirk relies on data supplied by Mr. Spock. They want to be involved in decision making, not just advising. 

The dearth of data scientists of this caliber has become a constraint on some sectors, forcing people to devise their own ways of generating and locating talent. Greylock Partners, a venture firm, has built its own specialized recruiting team to channel talent to the businesses in its portfolio. And after acquiring the big data firm Greenplum, EMC launched a data science and big data analytics training and certification program. 

Once companies snag a good data scientist, it's hard to hold onto him or her. LinkedIn lost Jonathan Goldman to Aster Data, which lost him to Level Up Analytics, a company Goldman co-founded.

A 191-meter-high Ferris wheel with 36 pods, each big enough to hold 40 gawking tourists, is to be built on New York's Staten Island, the New York City government announced today.

Ahhh. At long last, we New Yorkers have a mega-engineering project of our own.

It's been a long time since big things were built for the sake of bigness here, or anywhere else in the United States. Other countries, particularly in Asia, still have that soaring ambition, as this list of the tallest skyscrapers shows. But Ferris wheels are even more fun: Singapore is home to the tallest wheel, the 165-meter Singapore Flyer. Beijing was to have had a 208-meter wheel by 2008, but the finances went sour, and the project was shelved in 2010.

New York's wheel would provide not only a great ride but also a chance to see the Manhattan skyline —one more reason, besides the free Staten Island Ferry (which the wheel will abut), to visit the least-toured borough in the city. Construction is supposed to start in 2014 and end the following year. First, though, the project has to get through the city’s byzantine system for land-use approval.

New York Wheel LLC, the investing group behind the project, says it expects to spend US$250 million on the wheel and accompanying buildings. The contractor, Netherlands-based Starneth, built the 135-meter London Eye in 2000. A hydraulic system driven by electric pumps turns that wheel in 30 minutes; a ride on the New York Wheel is expected to take 38 minutes.

One of the main technical challenges to these things is just getting the parts of the structure into place. Starneth built the London Eye by barging in sections on the Thames, assembling them flat against the ground, and then using a jack system to elevate the 1,200-metric-ton wheel over the course of a day.