Enterprise architects from two early cloud adopters, a major medical supply vendor and an international real estate firm, shared lessons learned at the recent Consumerization of IT in the Enterprise conference.
This article spotlights their advice and anecdotes on cloud provisioning, getting user buy-in, best practices in a cloud pilot program and more. We'll hear from Tom Soderstrom, CTO at NASA's Jet Propulsion Labs (JPL); Brian Gillespie, enterprise architect at Cerner Corporation; and Cary Sylvester, vice president of technology, innovation and communications at real estate company Keller Williams.
The future of the enterprise software stack
In 2008, NASA's JPL Labs launched an enterprise-wide cloud initiative with about 5,000 people and an annual budget of about $1.5 billion dollars. The group is responsible for managing 19 active spacecrafts and 9 instruments across the solar system. JPL made an executive decision to rent servers rather than buy them. The goal was to replace a procurement screen with a provisioning screen.
The Linux, Apache, MySQL, PHP (LAMP) stack has powered the current generation of Web applications, but Soderstrom believes the next generation of IT architecture will be driven by business processes built on social, mobile, analytics and the cloud.
JPL used to budget new projects on an annual basis, "but we had to change that," Soderstrom said. "Instead of one humongous data center, you can experiment."
It is not just the cloud, but rather how it is used that determines security risks.
For the right use case, the cloud is amazing, Soderstrom said. For example, you can run 100 hours of computing on one machine or 100 machines for an hour and it would cast the same amount. Higher-performance computing is reachable, as well. "We can rent one of the top 500 supercomputers in the world for $35 per hour," he noted.
The landing of NASA's Curiosity Rover was one of JPL's biggest events, and JPL decided to use the cloud for the broadcast.
JPL had never done a large stream, and yet, two weeks before the landing, they used the cloud to spin up new infrastructure as required. As it got close to landing, they hit 3.6 million page views. As it landed, the numbers jumped to 14.1 million viewers.
"We were able to share the excitement with the world," Soderstrom said. The whole project ended up being 100 times less expensive than the last Rover project, even though they broadcasted to a larger audience.
Start experimenting with non-secure data
When evaluating cloud infrastructure offerings, it is important to separate fact from myth, Soderstrom said. For example, the questions about security are almost meaningless. It is not just the cloud, but rather how it is used that determines security risks.
One good practice is to segment your applications into different security needs. Then you can prioritize and move the apps with lowest security needs to the cloud. For example, JPL started tackling open data first, which allowed them to experiment with the cloud without creating risks in the event of a data leak, except perhaps embarrassment, Soderstrom said.
The next step is to evaluate highly sensitive data and work out any security risks – such as server hijacking or data loss or leakage -- before moving it to the cloud. "We found that we were spending as much money protecting cafeteria menus as spacecraft uplink commands," Soderstrom said.
While JPL doesn't have high security requirements for much of its data, it does have some requirements for protecting data related to International Traffic and Arms Regulation or for protecting the upload servers that control the spacecrafts.
For example, when JPL wanted to improve the infrastructure for communicating with remote vehicles, it turned to the cloud to reduce costs. The cloud ended up being 20 times faster and the cost was 10% less than the previous spacecraft control application that had been developed in-house.
Commoditize the cloud
Another myth about the cloud is that payment is as simple as swiping a credit card. For the enterprise, you need a contract, Soderstrom said. In JPL's case the best contracting plan involved creating a cloud computing commodity board, a strategy it had used previously for contracting water or electrical expenses.
The group meets once per quarter and includes the heads of IT, legal, security, finance and procurement, as well as customers. They look at the cloud and decide if it is a good prototype by getting a few billing cycles with the provider.
"This means we can keep evolving as the cloud community evolves," Soderstrom said. "This does not cost anything except getting people together once per quarter," he said.
Start with the carrot
When Cerner Corporation, a medical supply vendor, decided to adopt the cloud to improve their sales process, they realized that getting users on board would be tough. Many sales managers would maintain their own private spreadsheets, and were reluctant to share them, Gillespie said.
When they decided to bring this information into a centralized cloud application, they started working with a small team of users. Once other managers saw how the cloud could make their lives easier, other sales reps wanted to participate as well. This gave the IT department some leverage in setting policies, like the frequency of updates because the new users could see the value of the cloud.
"If we had just said, 'Go use this tool,' they would have balked," Gillespie said. "As we rolled it out and people could see it used, we had more leverage in setting policy."
Make search relevant
Building good search in the enterprise can be challenging, Gillespie noted. One of Cerner's biggest complaints was that users could not find things. Adding an enterprise search engine helped a bit, but there was not enough volume of searches to build a good index for representing what users were looking for.
To address this challenge, Gillespie recommended building a synonym matching index. For example, when a Cerner rep entered "Miami Kids CIO", he might be looking for someone at the Miami Children's Hospital. The challenge is that a health system, hospital or clinic might be separate businesses. To address this, they needed to build healthcare logic to make it easier to connect sales reps with what they were looking for.
Automating the social network
Gillespie said a good practice to keep in mind is to automate data entry as much as possible. Instead of using sales territory assignments, they can use the tool to see who is creating opportunities and quotes, and which applications they use.
This approach also makes it easier to create a social networking capability without adding much burden to the user, in terms of manual data entry. In Cerner's case, email was the preferred medium. It got approval to mine emails for the header information and to build up the social graph. At first, the CIO found this a little invasive, even though the tool did not look at email content.
The team built an algorithm that populates the social graph database automatically which made it easier to allow warm handoffs between sales reps. Also, the application could preemptively notify sales reps if a client had not been contacted in 45 days.
The sales reps also used a IP application to make calls, which made it easy to mine their call history and populate this information into the social graph. Down the road, Gillespie would like to do the same thing using the call logs from mobile providers.
This approach allowed the application to automatically generate information from about 36% of the contacts; the rest were a bit more challenging. Consideirng, Cerner turned to a cloud-based email marketing service normally used to retrieve email addresses from specific contacts. The cloud service was able to do a reverse lookup on many of the emails and automatically populate complete contact information.
Understand the user experience
Keller Williams International wanted to use the cloud to improve mobile access to information for agents in the field. Sylvester said they started with a desktop application but wanted to have a native experience and be able to leverage data and sensors on the phone.
They worked with a mobile development company to identify what information sales agents were using the most. These are the sorts of things that were designed to be accessible from a single click.
Sylvester recommended setting up a wall in your office to map out what users are doing on a day-to-day basis. It is also a good idea to sit behind users in order to understand the way they interact with the application on the mobile phone and tablet.
Pilot groups can also help to not only assess the application, but also assess the training requirements for getting new users up and running smoothly. By seeing where early users get stuck, they have a better idea on how to train new users. This approach has made adoption quicker and easier.