29 August 2007

Why "Good Enough" Is Good Enough

From BusinessWeek September 3, 2007

NEWS & INSIGHTS/Commentary
By Stephen Baker

Why "Good Enough" Is Good Enough
Imperfect technology greases innovation--and the whole marketplace

Say you have a crucial conference call in an hour and your phone goes dead. What do you do? A generation ago, this wasn't much of an issue, at least in the U.S. Phones in the days of the Bell monopoly were engineered to be "mission critical." You picked up one of those heavy receivers back then, and the dial tone was as prompt and reliable as water from the tap. It worked.

Yet these days, even as we pack global multimedia in our pockets, phone service sometimes seems to march backward. Andy Beal was one of 220 million subscribers to Skype, the cut-rate Internet telephony service owned by eBay (EBAY ), who saw the service go dark on Aug. 16. A software glitch kept it down for the next two days. Founder of the Raleigh (N.C.) Internet marketing consultancy Marketing Pilgrim, Beal learned that Skype was out an hour before clients were to call him from Holland. He had to message them in a hurry, telling them to call his tenuous backup: the cell phone. "It was embarrassing," he says. But at least the cell phone worked--which isn't always the case.

Are communications getting worse? Not by a long shot. We're surrounded by miraculous machines and services, most of them calibrated to a level software engineers have long called "good enough." In the right circumstances, good enough is great for the entire economy. A marketplace that's not hung up on fail-safe standards is open to risk and innovation, and drives down prices. Ever since the dawn of the PC--the archetype for a good-enough machine--inventors have been freer than ever to piece together and launch their visions. Some are brilliant, some are half-baked, many are a blend of the two. A precious few are up and running 99.999% of the time--Bell's old standard. But they cost far less to build.

The rise of good-enough technology raises different questions for do-it-yourselfers and major corporations alike. It's no longer whether we can afford a technology, but more often whether we can afford the disruption if and when it fails. Is it critical? Do we have backup in place? Many of us face this question every time we venture from our office with a cell phone. We don't have "one machine that works all the time," says Dave Morgan, chairman of Tacoda Inc., a New York advertising company. "We have lots of alternatives that work most of the time."

The upside of this sloppy status quo is enormous. Consider Andy Beal. He pays Skype about $60 a year, plus a couple cents for foreign calls. This gives him global telephony wherever he wanders with his laptop. He calls the service "seamless." He recently switched most of his office work--including e-mail, contacts, and calendar--to free Web services. This, of course, entails risk. In late July, an electrical outage in San Francisco brought some of the biggest sites, from Craigslist to Second Life, crashing down for 12 hours.

Beal's data reside on Google (GOOG ). The search giant is in fact an example of a major corporation that, like so many small fry, bets its business on good-enough technology. Google's data centers, the heart of the company's operations, consist of hundreds of thousands of commodity computers wired into a vast global network. These computers are little more reliable than yours or mine. Many die and are replaced every week. It could be that one of them at this very minute is issuing its dying blinks--and taking down Andy Beal's contact data with it. But if Google is working as designed, it links customers to another copy of those files or Web pages stored elsewhere on the network. Every computer has a legion of backups. Success, in the good-enough economy, means racing ahead even as the machines supporting us sputter and break.
I am fortunate to have the the name Steve. So when i remind myself to K.I.S.S, it means Keep It Simple Steve. But how simple can you make something and still achieve the required objectives? After a little thought, boiled the question down to the word "Precision". How "Precise" does something have to be?

Sometime later, was sitting in a tavern, famous for its Red Door, talking with a fellow IT guy about computer stuff. We each read several weekly "rags" and would exchange thoughts on what had "Legs" and what was "Flash". Asked him for his thoughts on how precise all this tech needs to be to work.

He answers, "Aristotle."

He informs me that i am couple thousand years behind and that one of the original founders of the first "SAP" had been down this road. He is brief but explains, sometimes "less is more." Maybe the time is right for a cell phone that only makes calls. I have one. It resides in the console of me truck, turned off.

Or here is a digital solution that does everything I want (or need) and nothing else to clutter its beauty. I use about 2% of what is in any of the operating systems and applications on the machines in the lab. Ninety plus some percent of available resources is not needed, nor wanted, but you get it regardless. Sometimes people may be asking for the wrong thing when they exclaim, "Why can it not just work?', when they should ask, "Why can i not get just what i need?" Sometimes more is less.

Until the next post,

Steve


Really Big Clusters Made From Lots of Small Servers

Is the real interest in "Greener" data centers to lower energy costs (good for the environment and bottom line) or is it that more efficient ways of computing require less energy? Maybe some of both?

From InfoQ:

Greener datacenters through Millicomputer clusters?

Posted by Johan Strandler on Aug 28, 2007 10:07 AM


One big problem with current large scale enterprise computing and data centers is power consumption, and a lot of effort is made in the industry to reduce the power need in current server platforms. Adrian Cockcroft is defining a new type of enterprise computing platform where he addresses this problem by defining a new type of computer: The Millicomputer - a computer that requires less than 1 Watt. The idea is to build enterprise servers out of commodity components from the battery powered mobile space. He presents a way to build an enterprise server using about 100 such Millicomputers in a cluster on a single 1U rack. This server only consumes less than 160W which is much less than comparable 1U rack enterprise servers of today. Cockcroft calls this disruptive innovation and he makes a prediction for 2010 that there could be a market for about 100,000 Milliclusters at $10K each, where each Millicluster packages 100 Millicomputers into an Enterprise Server.

The Millicomputer and MilliCluster hardware is developed as "Open Hardware", which means that the hardware design won't be owned by a single vendor. The Millicomputer is using LInux as the operating system and the hardware is based on a Freescale i.MX31 System-on-a-chip component using microSDHC flash memory. While the Millicomputer doesn't require much power by itself, external ethernet connections do. In order to save power, Cockcroft introduces the concept of "Enterprise MilliCluster", which allows 14 Millicomputers to be load balanced behind one Twin 1GB Ethernet external interface ethernet port by connecting by connecting them through a USB switch using Linux USBNet transport. The form factor to such a MilliCluster makes it possible to put 8 clusters plus a power unit on a single 1U rack, which consumes less than 160W - probably much less.

By comparing a MilliCluster based 1U server with Suns x4100 Operon and T1000 Niagara servers Cockcroft says:

"For the same 1U package size and similar cost per package power is much less than a Niagara, less than half of an Opteron system. Total RAM capacity is similar, the raw CPU GHz is double, worst case GHz per Watt is six times better than Opteron, three times better than Niagara. Flash storage is 1000x faster for both random and sequential IOPS.

Applications suitable to run on Millicomputers include:

Applications that can be broken into small chunks, small scale or horizontally scalable web workloads, legacy applications that used to run on 5 year old machines, graphical video caves and storage I/O intensive applications are the best candidates to run on Millicomputers."

Although in very early development, Millicomputing appears to be quite a paradigm shift; could this be the enterprise hardware platform of the (greener) future?

Until the next post,

Steve