Microsoft Embraces Scale

August 4, 2009

The year was 2002. A cash-rich, confused outfit paid me to write a report about Google’s database technology. In 2002, Google was a Web search company with some good buzz among the alleged wizards of Web search. Google did not have much to say when its executives gave talks. I recall an exchange between me and Larry Page at the Boston Search Engine Meeting in 1999. The topic? Truncation. Now that has real sizzle among the average Web surfer. I referenced an outfit called InQuire, which supported forward truncation. Mr. Page asserted that Google did not have to fool around with truncation. The arguments bored even those who were search experts at the Boston meeting.

I realized then that Google had some very specific methods, and those methods were not influenced by the received wisdom of search as practiced at Inktomi or Lycos, to name two big players in 2000. So I began my research looking for differences between what Google engineers were revealing in their research papers. I compiled a list of differences. I won’t reference my Google studies, because in today’s economic climate, few people are buying $400 studies of Google or much else for that matter.

I flipped through some of the archives I have on one of my back up devices. I did a search for the word “scale”, and I found that it was used frequently by Google engineers and also by Google managers. Scale was a big deal to Google from the days of BackRub, according to my notes. BackRub did not scale. Google, scion of BackRub, was engineered to scale.

The reason, evident to Messrs. Brin and Page in 1998, was that the problem with existing Web search systems was that the operators ran out of money for exotic hardware needed to keep pace with the two rapidly generating cells of search: traffic and new / changed content. The stroke of genius, as I have documented in my Google studies, was that Google tackled the engineering bottlenecks. Other search companies such as Lycos lived with the input output issues, the bottlenecks of hitting the disc for search results, and updating indexes by brute force methods. Not the Google.

Messrs. Brin and Page hired smart men and women whose job was “find a solution”. So engineers from Alta Vista, Bell Labs, Sun Microsystems, and other places where bright folks get jobs worked to solve these inherent problems. Without solutions, there was zero chance that Google could avoid the fate of the Excites, OpenText Web index, and dozens of other companies without a way to grow without consuming the equivalent of the gross domestic product for hardware, disc space, bandwidth, chillers, and network devices.

Google’s brilliance (yes, brilliance) was to resolve in a cost effective way the technical problems that were deal breakers for other search vendors. AltaVista was a pretty good search system but it was too costly to operate. When the Alpha computers were online, you could melt iron ore, so the air condition bill was a killer.

Keep in mind that Google has been working on resolving bottlenecks and plumbing problems for more than 11 years.

I read “Microsoft’s Point Man on Search—Satya Nadella—Speaks: It’s a Game of Scale” and I shook my head in disbelief. Google operates at scale, but scale is a consequence of Google’s solutions to getting results without choking a system with unnecessary disc reads. Scale is a consequence of using dirt cheap hardware that is mostly controlled by smart software interacting with the operating system and the demands users and processes make on the system. Scale is a consequence of figuring out how to get heat out of a rack of servers and replacing conventional uninterruptable power supplies with on motherboard batteries from Walgreen’s to reduce electrical demand, heat and cost. Scale comes from creating certain propriety bits of hardware AND software to squeeze efficiencies out of problems caused by physics of computer operation.

If you navigate to Google and poke around you will discover “Publications by Googlers”. I suggest that anyone interested in Google browse this list of publications. I have tried to read every Google paper, but as I age, I find I cannot keep up. The Googlers have increased their output of research into plumbing and other search arcana by a factor of 10 since I first began following Google’s technical innovations. Here’s one example to give you some context for my comments about Mr. Nadella’s comments, reported by All Things Digital; to wit: “Thwarting Virtual Bottlenecks in Multi-Bitrate Streaming Servers” by Bin Liu and Raju Rangaswami (academics_) and Zora Dimitrijevic (Googler). Yep, there it is in plain English—an innovation plus hard data that shows that Google’s system anticipates bottlenecks. Software makes decisions to avoid these “virtual bottlenecks.” Nice, right? The bottleneck imposed by the way computers operate and the laws of physics are identified BEFORE they take place. The Google system then changes its methods in order to eliminate the bottleneck. Think about that the next time you wait for Oracle to respond to a query across a terabyte set of data tables or you wait as SharePoint labors to generate a new index update. Google’s innovation is predictive analysis and automated intervention. This explains why it is sometimes difficult to explain why a particular Web page declined in a Google set of relevance ranked results. The system, not humans, is adapting.

I understand the frustration that many Google pundits, haters, and apologists express to me. But if you take the time to read Google’s public statements about what it is doing and how it engineers its systems, the Google is quite forthcoming. The problem, as I see it, has two parts. First, Googlers write for those who understand the world as Google does. Notice the language of the “Thwarting” paper. Have you thought about Multi-bitrate streaming servers in a YouTube.com type of environment. YouTube.com has lots of users, and streams a lot of content. The problems are that Google’s notion of clarity is show in the statement below:

equation

Second, very few people in the search business deal with the user loads that Google experiences. Looking up the location of one video and copying it from one computer to another is trivial. Delivering videos to a couple of million people at the same time is a different class of problem. So, why read the “Thwarting” paper? The situation described does not exist for most search companies or streaming media companies. The condition at Google is, by definition, an anomaly. Anomalies are not what make most information technology companies hearts go pitter patter more quickly. Google has to solve these problems or it is not Google. A company that is not Google does not have these Google problems. Therefore, Google solves problems that are irrelevant to 99 percent of the companies in the content processing game.

Back to Mr. Nadella. This comment sums up what I call the Microsoft Yahoo search challenge:

Nadella does acknowledge in the video interview here that Microsoft has has not been able to catch up with Google and talks about how that might now be possible.

I love the “might”. The thoughts the went through my mind when I worked through this multi media article from All Things Digital were:

  1. Microsoft had access to similar thinking about scale in 1999. Microsoft hired former AltaVista engineers, but the Microsoft approach to data centers is a bit like the US Navy’s approach to aircraft carriers. More new stuff has been put on a design that has remained unchanged for a long time. I have written about Microsoft’s “as is” architecture in the Web log with snapshots of the approach at three points in time
  2. Google has been unchallenged in search for 11 years. Google has an “as is” infrastructure capable of supporting more than 2,200 queries per second as well as handling the other modest tasks such as YouTube.com, advertising, maps, and enterprise applications. In 2002, Google had not figured out how to handle high load reads and writes because Google focused on eliminating disc reads and gating writes. Google solved that problem years ago.
  3. Microsoft has to integrate the Yahoo craziness into the Microsoft “as is”, aircraft carrier approach to data centers. The affection for Microsoft server products is strong, but adapting to Yahoo search innovations will require some expensive, time consuming, and costly engineering.

In short, I am delighted that Mr. Nadella has embraced scale. Google is becoming more like a tortoise, but I think there was a fable about the race between the tortoise and the hare. Google’s reflexes are slowing. The company has a truck load of legal problems. New competitors like Collecta.com are running circles around Googzilla. Nevertheless, Microsoft has to figure out the Google problem before the “going Google” campaign bleeds revenue and profits from Microsoft’s pivotal business segments.

My hunch is that Microsoft will run out of cash before dealing the GOOG a disabling blow.

Stephen Arnold, August 4, 2009

Comments

One Response to “Microsoft Embraces Scale”

  1. Microsoft Gets an F from Professor Google for Scale Paper : Beyond Search on August 16th, 2009 3:03 am

    […] wrote a short post here about Microsoft’s suggestion that Google has gone off track with its engineering for petascale […]

  • Archives

  • Recent Posts

  • Meta