Stephen E. Arnold's Blog

Google has shifted from solving problems in distributed, massively parallel computing to developing next-generation cloud-centric applications. Google can, with the deployment of software, deliver global services that other companies cannot match in terms of speed of deployment, operation, and enhancement.

Cloud computing has become commonplace. Amazon has pumped steroids into the Amazon Web Services product line. Microsoft executives have been providing forecasts of a bold new service offering. Other vendors blasting off from mother earth to loftier realms include IBM, Intel, Rackspace, and other big name firms.

One of the most interesting documents I have read in months is a forthcoming technical paper from Microsoft’s Albert Greenberg, Paranta Lahiri, David Maltz, Parveen Patel, and Sudipta Sengupta. The paper is available from the ACM as document 978-1-60558-181-1/08/08. I have a hard copy in my hand, and I can’t locate a valid link to an online version. The ACM or a for fee database may help you get this document. In a nutshell, “Towards a Next Generation Data Center Architecture: Scalability and Commoditization” explains some of the technical innovations Microsoft is implementing to handle cloud-based, high-demand, high-availability applications. Some of the information in the paper surprised me. The innovations provide a good indication of the problems Microsoft faced in its older, pre-2008 data centers. It was clear to me that Microsoft is making progress, and some of the methods echo actions Google took as long ago as 1998.

What put the Amazon and Microsoft cloud innovations into sharp relief for me was US2008/0262828 “Encoding and Adaptive Scalable Accessing of Distributed Models.” You can download a copy of this document from the easy-to-use USPTO system. Start here to obtain the full text and diagrams for this patent application. Keep in mind that a patent application does not mean that Google has or will implement the systems and methods disclosed. What the patent application provides is a peep hole through which we can look at some of the thinking that Google is doing with regard to a particular technical issue. The peep hole may be small, but what I saw when I read the document and reviewed the drawings last night (October 24, 2008) sparked my thinking.

Before offering my opinion, let’s look at the abstract for this invention, filed in February 2006 in a provisional application. Keep in mind that we are looking in the rear view mirror here, not at where Google might be today. This historical benchmark is significant when you compare what Amazon and Microsoft are doing to deal with the cloud computing revolution that is gaining momentum. Here’s Google’s summary of the invention:

Systems, methods, and apparatus for accessing distributed models in automated machine processing, including using large language models in machine translation, speech recognition and other applications.

In typical Google style, there’s a certain economy to the description of an invention involving such technical luminaries as Jeff Dean and 12 other Googlers. The focus of the invention is on-the-fly machine translation. However, the inventors make it clear that the precepts of this invention can be applied to other applications as well. As you may know, Google has expanded its online translation capability in the last few months. If you have not explored this service, navigate to http://translate.google.com and try out the system.

The claims for this patent document are somewhat more specific. I can’t run through the 91 claims in this patent document. I can highlight one, and I will leave review of the other 90 to you. Claim 5 asserted:

The system of claim 4, wherein: the translation server comprises: a plurality of segment translation servers each operable to communicate with the translation model server, the language model servers and replica servers, each segment translation server operable to translate one segment of the source text into the target language, a translation front end to receive the source text and to divide the source text into a plurality of segments in the source language, and a load balancing module in communication with the translation front end to receive the segments of the source text and operable to distribute the segments to the segments to the segment translation servers for translation based on work load at the segment translation servers, the load balancing module further operable to direct translated segments in the target language from the segment translation servers to the translation front end.

The claim makes reasonably clear the basic nesting architecture of Google’s architecture. What impressed me is that this patent document, like other recent Google applications, makes use of an infrastructure as platform. The computational and input output tasks are simply not an issue. Google pretty clearly feels it has the horsepower to handle ad hoc translation in real time without worrying about how data are shoved around within the system. As a result, higher order applications that were impossible even for certain large government agencies can be made available without much foot dragging. I find this remarkable.

This patent document, if Google is doing what the inventors appear to be saying, is significantly different from the innovations I just mentioned from such competitors as Amazon and Microsoft. Google in my opinion is making it clear that it has a multi-year lead in cloud computing.

The thoughts that I noted as I worked thorough the 38 pages of small print in this patent document were:

1. Google has shifted from solving problems in distributed, massively parallel computing to developing next-generation cloud-centric applications. Machine translation in real time for a global audience for free means heavy demand. This invention essentially said to me, “No problem.”
2. Google’s infrastructure will become more capable as Google deploys new CPUs and faster storage devices. Google, therefore, can use its commodity approach to hardware and experience significant performance gains without spending for exotic gizmos or try to hack around bottlenecks such as those identified in the Microsoft paper referenced above.
3. Google can, with the deployment of software, deliver global services that other companies cannot match in terms of speed of deployment, operation, and enhancement.

I may be wrong and I often am but I think Google is not content with its present lead over its rivals. I think this patent document is an indication that Google can put its foot on the gas pedal at any time and operate in a dimension that other companies cannot. Do you agree? Disagree? Let me learn where I am off base. Your view is important because I am finishing a write up for Infonortics about Google and publishing. Help me think straight. I even invite Cyrus to chime in. The drawings in this patent application are among Google’s best that I have seen.