• Home
• Articles Archive
• Features List

• Media Pack
• Contact Us
• Subscribe

Wired for service

From STORAGE Magazine Vol 7, Issue 2 - March/April 2007

Pundits are already singing the praises of utility computing - making services available on demand by separating application services from the IT infrastructure that supports them. Yet turning the theory into practice is not without its challenges

International IT services company ATOS Origin is one organisation that has made real progress towards the challenging process of making services available on demand by separating application services from the IT infrastructure that supports them. The company's senior consultant Adil Tahiri has been telling Storage magazine how his organisation is moving towards this goal across its UK operations, helping it become more responsive to customers and staying keenly cost efficient at the same time.

"Many IT directors proudly tout their data centres as the backbone of their businesses," states Tahiri. "In our case, it supports not just us, but our customers as well. Some of the UK's best-known organisations entrust their data to us." In fact, ATOS Origin - with revenues of more than 5 billion Euros and 47,000 staff - has clients that include major public sector organisations and leading companies such as ABN Amro, ING, Lloyds TSB, National Express and Vodafone. So the undertaking carries with it huge responsibility.

"Regardless of whether the customer is internal or external, for our business to continue growing, we have service level agreements [SLAs], which we must meet, while at the same time keeping costs well under control.

"As with most organisations, our data centres were originally based on a siloed view of the world, in which storage, processing and network infrastructure support specific processes independently of each other. This approach made service management and delivery very simple, but caused headaches in other areas."

Processor and storage utilisation hardly reached 50%, so a significant proportion of the company's resources were unused. "Add together the capex [capital expenditure] and opex [operational expenditure] on keeping these silos of disparate resources running, not to mention the cost and time associated with provisioning additional resources for customers, and you very soon reach dizzying sums," he adds.

At the same time, the company was bidding for contracts, all of whose technical requirements were broadly similar. Rather than implement discrete platforms for each customer, ATOS Origin reasoned, it could build a data centre network architecture that presented storage, processor and network as shared networked resources to the applications, for customers to draw on as necessary.

"We developed a new blueprint for our data centre," explains Tahiri, "based on a common 'distribution layer', comprised of the actual server, storage and network hardware, and an 'access layer', tailored to each customer, which ensured their applications and data were always available.

"This service-oriented approach to the data centre would make it much easier for us to respond quickly to customers - adding more capacity or plugging in more services, for example. However, compliance, security and service delivery were key concerns: could our approach compromise our customers' applications or data? Would our ability to meet SLA deliverables be impacted, if a small number of customers experienced 'spikes' of high demand?"

End-to-end environment
As Tahiri points out, there are a large number of vendors converging on the 'utility computing' space, offering alternative approaches with tools that support load-balancing, shared resources, service management, etc.

"Meanwhile, a few, such as Cisco Systems, provide the entire solution off the shelf. Rather than backing one approach over the other out of hand, we convened two working groups to evaluate them in parallel. Whilst alternative approaches might offer a more customised solution, we, in the end, opted for an end-to-end Cisco environment - and not just because it was our existing network partner.

"We needed reassurance that our vendors would be able to support us five or ten years down the road, not just this year and next. Cisco's view of the future was a close match with our own, backed by a proven track record of investment protection and solid integrated data centre technology roadmap."

These considerations, he adds, are vital for any organisation considering utility computing and aren't easy to map out with multiple vendors. "Will the same network architecture I am deploying today support storage virtualisation and data migration technologies I want to deploy in the future? How will the network support the rack optimised servers I am currently standardised on and the blade servers I may choose to deploy next year? Will I be able to optimise application performance levels for any new customer applications, based on the infrastructure I deploy today?

“How will high-performance computing or grid computing impact this framework, should my customer decide it is a requirement at some stage in the future? In our view, the potential complexity of these questions made a single-vendor solution the best and safest option."

Standardise and deliver
Having previously standardised all of its data centre processes - from SAN traffic to application data - on IP, the company had already laid the foundations for the future. The next stage involved a sophisticated switching infrastructure to keep the thousands of data streams separate, whilst actually drawing on shared computing resources.

"Given our concerns over security, this step was especially important, as it enabled us to roll out multiple security policies over the same underlying infrastructure. We completed our distribution layer with a network of switches upgraded to provide firewalls, load balancing and content switching. Rather than bringing the entire facility offline and getting out the heavy lifting gear for a wholesale upgrade, we were able to make straightforward modular upgrades to our existing Cisco Catalyst 6500 switches. This step alone saved considerable sums.

"Another series of switches made up the access layer, governing access to distribution layer resources in line with security policies and SLAs. With the distribution layer effectively acting as a 'service' to the access layer, we are able to provision more and more services - further web service functionality or content caching, for example - with ease. In our previous environment, we would typically have needed to order any new hardware, and spent time configuring and testing the device before putting it into production.

"This approach has greatly simplified our hardware footprint, making it easier to respond to customer requests and taking much of the detective work out of service management. What's more, our two-tier strategy, and Cisco's own approach to investment protection, has helped us keep costs - capex and opex - well under control."

Tahiri says the foundation that has been laid will serve ATOS Origin well into the future.

"We are now ready to roll out fully-blown server/storage virtualisation and end-to-end service provisioning at the click of a mouse. What is more, migrating services from one platform to another will take a fraction of the time it did previously, so we expect to slash downtime. As our business depends on our planning ahead of our clients' demands, we believe our data centre network has left us well equipped for the future." ST