Case Study: A More Resilient Application Infrastructure

Company taps on-demand delivery model to enhance retail management application

Jeff Almoney knows a thing or two about scalable, highly available computing technologies. During a stint as director of advanced information technology with the Pennsylvania State University, he oversaw the implementation of one of the first IBM Corp. RS/6000 SP2 clusters—a system with serial number 3. Serial numbers 1 and 2, he says, were deployed at “Cornell University and some lab, somewhere.” To this day, Almoney remains on a first-name basis with Irving Wladawsky-Berger, vice-president of technology strategy for IBM, and the visionary behind SP2 and other technologies.

These days, Almoney is CTO of Reynolds & Reynolds Co., a provider of customer service and training solutions for several industries, including almost 40 percent of North American automotive retailers. Reynolds typically provides clients with some of the hardware, software, and services they need to support their everyday business operations. Reynolds has been proud of its ability to deliver scalable and reliable applications and services to more than 20,000 customers.

Scalability Shortcomings

The rub, Almoney says, was that his employer’s Intel-based server infrastructure had long ceased to be either scalable or reliable. There was a good reason for this: Reynolds’ “infrastructure” wasn’t entirely in-house; it included an “extra-structure” of systems scattered across customer sites.

At first, Reynolds tried to mitigate this situation by transitioning to an ASP-type service model and centralizing on rack-mounted servers, but the approach led to additional problems. Reynolds’ ASP server farm didn’t scale well and was inherently difficult to manage. Just as important, even with the ASP model, the company’s customers still had to assume some of the management burden.

The issue, as Almoney dryly notes, was that few automotive dealers exist solely to manage their IT systems; but with vanilla Intel-based servers, some dealers had become part-time IT administrators by default.

What Reynolds needed, Almoney determined, was an adaptive infrastructure that was at once more resilient and more reliable. At a minimum, he figured, Reynolds would need to centralize the management and administration of its applications and services to the greatest degree possible. This meant offloading as much of the day-to-day hosting and administration of customers’ systems and applications as was possible. In other words, Reynolds needed a pipedream.

“What we were looking for was a way to upgrade the systems that the dealers had access to, but, ideally, we would also be able to shift the burden of managing these [systems] back to us, too,” he says. “But just as important was the scalability and reliability of the platform, and also the manageability—ideally, we were looking for autonomic capabilities, especially self-healing.”

Answer to a Pipedream

The answer to Almoney’s prayers came in a small form factor: a blade server. Reynolds & Reynolds tapped a next-generation infrastructure based on IBM’s BladeCenter technology, IBM Director blade management system, IBM’s Tivoli management offerings, and IBM TotalStorage arrays.

Reynolds’ blade systems run a mixture of Linux and Windows systems and are the culmination of a massive consolidation of the company’s former (highly distributed) server infrastructure.

In addition, the company decided to build an entirely new retail management application, based on an on-demand delivery model, that taps some of the autonomic and self-management capabilities provided by IBM’s Director BladeCenter management software and Tivoli management products.

One result, says, Almoney, is increased resiliency: If a system or component fails, BladeCenter automatically detects the failure and issues a notification without requiring human intervention. On top of that, BladeCenter—and, more specifically, IBM Tivoli management—helps to automate Reynolds’ backup and disaster recovery operations. This is a boon to customers, to say the least, says Almoney, because it helps to minimize the extent to which they’re forced to wear the IT manager’s hat.

“The beauty of the BladeCenter implementation is the complete remote control of the system. We have a completely remote control [environment]. There’s no physical attendance of our system. There are people available to do remote hand’s-on support, but the only reason there are hands in the system is to do repairs.”

It’s for this reason, says Almoney, that the BladeCenter paradigm is so very compelling. “So I think you can expect to make your life easier because you’ve got an infrastructure that has complete remote control, so the next piece is making sure you build it in a scalable way, so making sure you have a storage infrastructure that can support it.”

Icing on the Cake

There are other benefits, too. Increased resiliency translates, to some extent, into improved predictability. Improved predictability, Almoney says, correlates with better customer service. “You want your systems to be as reliable as possible. We certainly in our On Demand environment will give you an SLA, and because of that we want to give you a real enterprise-class delivery of service.”

BladeCenter is a highly centralized offering, Almoney notes, and to a large extent benefits from technology trickle-down from the rest of IBM’s stack. In this respect, he says, it reminds him of the best of what the erstwhile RS/6000 SP2 systems had to offer (in terms of scalability and integrated manageability) with autonomic and improved self-management capabilities thrown in to boot.

“The RS/6000 SP was about having lots of processors that can work on the same problem, so it had the ability to do message-passing and parallel processing. Not that you can’t do the same thing on the BladeCenter. In fact, the latest BladeCenters have InfiniBand,” he concludes.

So how does Reynolds & Reynolds new BladeCenter implementation measure up in the most important of metrics—ROI?

Almoney says it’s difficult to affix a price to what amounts to a wholesale transformation of its existing infrastructure. “For us, it’s difficult to just pull that piece out. We really run a better, higher-class system, and more expensive hardware is used in the centralized environment than we would ship to individual dealers,” he concludes.

“The ROI is a bit rough of a calculation. It’s difficult to put a price on the absolutely much better level of service we can offer customers [and] the way we’ve really reduced customer headaches.”

comments powered by Disqus