Governance, modelling unravel Coles-Myer's application 'hairball'

Just how complicated, however, wasn't clear even to the more than 1,100 IT staff working across the Coles-Myer group until the company set about a major systems refresh and development effort that forced it to take stock of more than 25 years' worth of organic systems growth.

Like a house of cards, those systems were working together to deliver the systems Coles-Myer needed -- but even small changes risked bringing the whole thing down. The challenge: with so many people involved in the systems and no single person understanding how the whole puzzle fit together, the company had no idea of just what were the interrelationships between the systems.

This situation presented a major problem when Coles-Myer embarked on a five-year reinvention of its systems, with the goal of moving away from aging and expensive legacy systems to a more flexible, capable environment. One major problem: technology at each of the company's subsidiary brands was managed more or less independently, a strategy that compounded the difficulties of moving towards a single J2EE (Java 2 Enterprise Edition) based service oriented architecture (SOA).

Map of the retail heart
From the early days of planning, it was clear to all involved that the effort would require a major mapping effort to ascertain just what was actually out there and running. The Metis interface modelling tool, from Norwegian company Computas, was eventually chosen as the base for the construction of a comprehensive road map that would label what Chris Tisseverasinghe, enterprise architecture manager with Coles-Myer, termed a "hairball" while presenting the solution to a meeting of the Enterprise Java Victoria user group in Melbourne this week.

"We had lots and lots of interfaces, and when one of those fell over in the middle of the night people would be running around," he explained. "They didn't understand what the impact might be, and the knowledge [about how to fix problems] was spread out throughout the company."

Despite the massive complexity of its systems, a slowly-slowly approach eventually helped the company's technical staff step through the modelling process until it had what was seen to be a comprehensive chart of its myriad application interfaces. Like a map of darkest Africa, that model exposed previously hidden details of complexities and interdependencies.

Such a map proved to be of major benefit when Coles-Myer made the decision to relocate hundreds of systems from its Hoddle Street, Melbourne data centre to a Hewlett-Packard-run co-location facility. Many of the systems in the Hoddle Street site predated the people maintaining them, with little idea of what was actually connected where. Even worse, hardware-focused data centre staff could offer little insight as to which business processes each server pertained -- so the impact of unplugging them was anybody's guess.

Since the move had to be conducted without causing even a second of downtime for the company's 24x7 business, the Metis map proved crucial in helping migration planners identify banks of servers by the application they affected, then move those servers to the new site without impacting application availability.

"We had to pick up boxes that had been there for 25 years, and no one knew what picking up one application would do to the business," said Tisseverasinghe. "By putting together models, we moved hundreds and hundreds of boxes in a six-month period without the business having an interruption. Impact analysis has proved to be really powerful, because we have all that information about our systems in the one place."

That analysis was also well received by helpdesk staff, who are tasked with resolving system problems but had been operating in the dark to match application problems with the systems that caused them. With a clear correlation between the two issues, helpdesk staff could be more proactive about noting server outages and contacting appropriate staff to let them know what applications would be affected.

Developers, however, were ultimately the biggest beneficiaries of the Metis modelling, since it guided their massive shift from Coles-Myer's largely inhouse developed systems towards a more consistent, industry-standard architecture based on IBM WebSphere and related applications providing functions like identity management, merchandising systems, loyalty programs, finance, HR, supply chain management, in-store stock applications, and so on.

Whereas systems in the previous environment were highly fragmented between brands, the new environment is seeking to unify that with a consistent XML-based data model. That has already led to big savings for Coles-Myer: after implementing the supermarkets data model, for example, the company was able to quickly bring Liquorland stores online by working their transactions into the same model. The savings, Tisseverasinghe said, were over AU$1 million just from that reusability alone.

Governance en masse
Because it has helped guide those major rollouts, use of the Metis tool ultimately provided benefits at every level of the IT organisation, acting as an effective governance tool for managing manage the company's massive architectural change. Having clear governance model was particularly useful in guiding solutions architects, who Tisseverasinghe said were "my eyes and ears on the ground, and [the ones that] effectively own the technical outcome for that project."

Even they need control, however: "if you [normally] let them loose they'll do what they want and walk away," he warned. "It's easy to put the projects on the ground, but unless you have a strong governance process, projects will do what they want. We have very clear principles and guidelines and responsibilities to keep systems architects and everyone else on the straight and narrow. They key thing is that these tools and architects can't stay in the back room; successful architects foster communication among wider audiences."

Ultimately, the only inhouse developed system will be the company's new portal, with everything based on J2EE but the .Net-based point-of-sale front end. The company's SOA is not fully in place yet, but the new system will eventually take an SOA approach that will be mapped by the continually updated Metis model.

As new systems are progressively rolled out during the continuing project, the Metis model is also helping tackle the even more daunting process of actually disconnecting old applications once and for all. Perhaps because of a philosophical attachment to what was working, or perhaps just because it's more exciting to implement new systems, many workers are usually reluctant to guide decommissioning -- but the new model has eased the process by guiding the unravelling of the hairball one strand at a time.

Automating many of its key systems will take at least three years' more work, but deployment of a clear governance model, linked to the business applications that technology serves, has proved to be invaluable during all of the reinvention of Coles-Myer's IT.

"Impact analysis has been a very powerful tool, because we have all that information in one place and we can manage the change over time," Tisseverasinghe said.

"The biggest pushback we've had has been from project managers and people who don't want to worry about the long-term game. But we've had a very strong implementation methodology, used a lot of best practices and used our own end-to-end processes. In the past, IT just responded to service requests, but now we've turned that around and said [to the business] we will tell you how we do it. We won't compromise on where we're going."