Case study: Clustering the penguins

By on
Case study: Clustering the penguins

The humble personal computer has a well-deserved reputation as a giant-killer. After all, it was the virus-like spread of PCs which spelled the beginning of the end for the mainframe dinosaurs.

Because the first low-cost servers emerged from high-end PCs being asked to look after more than one user at once, it took quite a few of the diminutive workhorses to match the capacity of the mini-computers they were slowly but surely replacing.

At first, this was heralded as a good thing – if one of these PC-based servers died, you’d have others still running. Of course, it wasn’t quite as easy to transfer users and applications from failed servers to working servers, but the theory of redundancy helped systems administrators sleep more peacefully.

Eventually, the little PC-servers evolved until they no longer looked anything like their PC-ancestors, instead becoming fire-breathing mini-dinosaurs with as much if not more capacity as their old mainframe nemesis. And they became just as hard to manage.

Systems managers cried out for fewer items of hardware to manage, and operating systems which made it easier to take advantage of the available power while ignoring the failed pieces of the puzzle.

Hardware vendors responded with ‘blade’ servers – simple plug-in servers which use a shared source of power and connectivity, and can be quickly swapped out and replaced should they fail. All that was required to complete the puzzle was the right kind of operating system, and both Microsoft’s Windows and various flavours of Unix offered the solution in the form of clustering.

Clustering as a technology dates back at least twenty years, but making it affordable is a recent phenomenon, and has been helped along enormously by the open source Linux operating system.

There were, and still are, many a heated argument about whether the price of the operating system really matters, when the hardware cost is quite high. But the advent of blade servers, and now clustering, suddenly elevates the price of multiple software licences to a significant budgetary consideration, as international payments and interactive transaction processing company, Dialect Payment Technologies soon discovered.

Dialect’s client base includes two of the top three global card companies and nearly 40 banks from around the world for whom the company provides Card Not Present payment processing services and also supports the electronic payments needs of their customers through consulting, integration, processing and support services.

In 2006 the company handled almost 55 million transactions. “Our business is critical to the success of our clients’ and is both demanding and competitive, and we simply can’t afford for our technology to let us – or our global client base – down,” says Francis Cox, COO for Dialect Payment Technologies.

When Cox was setting up the company’s server cluster, he evaluated the option of continuing with Microsoft’s Windows Server solution and concluded he would be paying 30 percent more for software licences if he took that path instead of using Linux, and would end up paying an extra $250,000 just for the right to use the software.

A figure of that magnitude would grab the attention of even a very well-heeled company. “When you’re talking about a quarter of a million dollar saving in licence costs, there is a very clear difference between whether we can be competitive or not,” says Cox.

However, the decision wasn’t purely cost driven, because Cox had determined that several java-based applications provided the best solution for his company’s processing requirements.

“Java-based applications and middleware products like Apache and Tomcat are clearly designed and built for Linux, and so our decision to move to Linux was largely driven by what best suited our applications,” says Cox. “As an organisation, we were also keen to move more towards an open standards environment.”

Since early 2005, Dialect has been running a mix of Red Hat Enterprise Linux versions 3 and 4 in a Hewlett Packard server cluster made up of 15 DL380 G4 servers with a storage area network backend.

Today, Dialect has 10 servers running Red Hat Cluster Suite to provide the 24x7 high availability the business requires, and Cox is looking to the future with plans to trial Xen virtualisation technology, as part of Red Hat Enterprise Linux 5, in 2007.

The company expects that utilising Xen will give them the ability to run multiple virtual operating systems on one physical server, reducing hardware and management costs even further.

Cox is also happy with the increased reliability provided by the Linux cluster. “Our previous published availability levels were at 99.5 percent – but from day one with Red Hat Enterprise Linux that extended to 99.7 percent, and today we are in a position where we are greatly exceeding our SLAs,” says Cox.

Saving money while making money and keeping the customers happy – just what every business strives to achieve, and the open source software revolution has helped Dialect attain that desirable status.
Tags:

Most Read Articles

Log In

Username:
Password:
|  Forgot your password?