3M Uses Azure For Image Analysis Service

8 Windows 8 Apps Under $25

When Bill Smyth wanted to launch a new business at 3M two-and-a-half years ago, he faced the daunting prospect of plunging into the $29.9 billion company's capital budgeting process. Or, he could circumvent that process and go straight to Microsoft's Azure cloud.

As a 14-year veteran at 3M with product responsibilities, he had been through the capital budget process several times. "I've got the scars on my back to prove it," he said in an interview.

Applying for capital funds puts an executive "into the competitive environment of capital money allocation," with repetitive meetings and discussions over how to divide up 3M's $1 billion capital budget. He only needed a few servers, but to get them, "I would be competing with executives building $100 million coating factories," he recalled. The budget process is the same whether someone needs three servers or a whole factory. And even if he received an allocation in one round of discussions, he could lose it in the next when a higher priority project was shown to be over budget and coming up short.

"I had a million things I knew I had to do" to launch 3M's Visual Attention Service (VAS), an image analysis service that advises clients which parts of their images or video will gain the greatest share of audience attention. Smyth only needed a handful of servers; going to the cloud gave him those servers quickly, along with one less thing to do.

"The concept of avoiding the countless budget meetings versus firing up more instances. It was a no brainer for me," he said.

[ Want to learn more about Microsoft's Windows Azure cloud services? See Microsoft Azure Public Cloud Matches Amazon Prices. ]

In another sense, Smyth also was circumventing IT as well as the capital budget. Rather than relying on on-premises servers, he was going outside the data center, and that posed data management challenges. But he had an internal ally, the 3M Software, Electronics and Mechanical Systems Lab, one of the central labs that serve the whole company. It was SEMS that had the neuroscience expertise to establish how the human brain perceives light and images. "SEMS played a key role in much of our software-related decision making," recalled Smyth. "They proposed using Azure."

Smyth was in a hurry to get his systems up and running because 3M realized two-and-a-half years ago that one of its underutilized, core competencies -- the neuroscience of image perception -- could be converted into a digital service. Two programmers from the lab became part of Smyth's five-person development team and were responsible for expressing the findings of the research in algorithms. And those algorithms became the foundation of his unit's VAS, which launched with Smyth as global business manager.

One early recognizer of the value of the service was Adobe, which offered the service through its Photoshop marketplace in 2010. The service was termed effective, if pricey, in a review by InformationWeek's sister publication CRN in April 2011.

Smyth felt sure many parties, from advertising agencies and marketing units to independent video makers, would want to use a service that could tell them what parts of their images were going to attract the most interest. Two-and-a-half years later, he's been proven right. VAS has logged 8,000 users, including 10 of the leading advertisers in the U.S. It turned out that fast food restaurants wanted to analyze the imagery on their digital menu boards to ensure they attracted customer attention to the right places. Television advertisers used it to figure out what parts of their commercials were attention getters. The same applies to makers of banner ads and display ads.

"Predicting the volume of traffic in the early days was difficult," he recalled. He knew he needed a compute infrastructure that scaled upward easily, in case growth took off. But he didn't want to over-invest as word was first getting out on the availability of the service. Smyth realized Azure fit his flexible scalability requirement also. The initial three virtual servers grew into a 12-server set of extra-large instances. That is, each virtual server has 14 GBs of RAM and eight virtual cores.

But these decisions weren't made entirely in the absence of IT, Smyth said. On the contrary, he kept the staff informed and solicited its support. In part, that's the culture at 3M. "People here like to work with and understand new technologies. IT knew what we were doing. They were interested in us as an early adopter and wanted to help," he said.

That help would be needed when it came to keeping VAS data secure. As companies started using the service, "they were developing things like new marketing materials and new packaging, [content] they view as sensitive company secrets."

LinkedIn: 10 Important Changes

Before using VAS, customers would conduct a security review of the service. Fortunately, 3M has its own security review process and made sure VAS complied with it. As name brands came to VAS, "we had our corporate IT data security people work with their data security people to help us get approved," he said. Before the service launched, Microsoft's Azure was subjected to a 3M IT security review and it passed as well.

There was another major reason 3M chose the Azure cloud. Smyth had a limited development staff: Two developers in the SEMS Lab produced the neuroscience-based algorithms and three developers in his business unit produced the user interface and business logic. To do so, they used technologies already approved for new apps at 3M: Microsoft's Visual Studio development tools, C# language and Team Foundation Server collaborative development platform. 3M is basically a Microsoft .Net shop, Smyth noted.

The easiest way to get the team to work together was to develop it on Azure. Once produced, other parts of the company involved in visual products -- 3M produces reflective paints, reflective signage and coatings for LCD screens to make them more usable, among other things -- could work with the neuroscience-based algorithms on Azure as well. So Visual Attention Service was developed, tested and staged on Azure, as well as running its production systems there.

3M also uses Microsoft's SQL Server database system internally; Azure offers a compatible cloud system, Azure SQL, that the VAS system uses, said Smyth. "Pretty much everything that the global, Visual Attention Service business uses is available in the cloud," he said. "If you're a Microsoft shop, it's just a straightforward process to go to the [Azure] cloud."

The needs of the 3M application are in the process of changing. Single image analysis is a compute-intensive process; the first clients came to VAS to analyze one image at a time. For the last six months, 3M has been quietly offering VAS for video as well, and use of that process helped push 3M's server total to 12. One second of video often consists of 30 images, several of them requiring the compute-intensive analysis process. He's knows if VAS needs to move beyond 12 servers, Microsoft will be happy to accommodate him.

At the same time, Smyth sometimes weighs the fact that there are alternatives available. The VAS service jumped onto Azure while it was still a beta service. "The documentation was decent, but there wasn't much of a user community. In some cases, we were pretty much on our own to figure out how do some things."

Since April 16, Azure has been generally available as infrastructure-as-a-service as well as a development platform, and more users appear on it each day. However, rival Amazon Web Services has a large user community that formed during the past seven years that the Amazon service has been available. The lack of a vigorous Azure user community to consult with "probably hurt," he conceded. "Could we have moved faster on Amazon? I really can't say." It was vital to use the same tools and technologies in the cloud as 3M did internally, he said.

The service suffered an outage on Feb. 29, 2012, the day that Azure security certificates failed to recognize that the quadrennial Leap Day had arrived. The lack of a Feb. 29 date for the certificates prompted a false reading by cluster governors of a hardware failure. The cascading readings caused a spreading failure throughout Azure that stretched across an eight to 10 hour period. Other than that, there have been "a couple of short-term glitches [on Azure] that were not an issue with customers," he said.

And so far, renting Azure instances still doesn't require a capital budget or attendance at a long set of meetings. "It's a local decision that I can make independently," he noted. That's a good thing, because he believes video analysis, now generally available, will generate greater VAS usage and drive his business forward. The impact of video on the Internet is becoming a factor in website and application success. And as more video makers need to use image analysis, Smyth has no intention of cutting his ties to the highly scalable Azure.