Killer searching and killer server savings, what else but Linux could give them all they want in a smooth landing into first-class e-commerce.

From the start, server technology has been the big competitive weapon for Orbitz. The Orbitz search engine is used to please customers with ample options shown for fares and flights. The Orbitz ticket booking system called “Supplier Link” is used to win the hearts of airline suppliers because it is a direct reservation system that bypasses fee-charging global distribution systems (GDS) to book directly with airline ticketing systems.

Not surprisingly, Linux plays a big part of the Orbitz success story. Linux is being used in a server-farm strategy within the company’s production environment. In an industry with daily reports of airline layoffs, tattered customer frills, ten-dollar sandwiches, and an ailing hospitality service riding its frayed coat-tails, anything that brings a competitive edge is welcome, and that goes double in the midst of an airline and hotel booking slump.

The only high-flyer in the travel industry has been the online sector, which has taken on a life of its own. Nearly 60% of adult Web surfers have shopped for travel-related services on the Internet in the last six months, according to Nielsen//NetRatings

The contrast between traditional industry sales and on-line sales has been documented by research firms like Connecticut-based PhoCusWright Research. They have found that online travel sales grew 47% in 2002, even while overall travel industry sales fell 5%. They also say that online travel has flourished to become one of America’s largest businesses—closing in on $30B+ annually.

Orbitz founders saw early on that the online travel marketplace, if worth billions, would be worth billions more in coming years. As soon as they launched in 2001, they started talking about how they would have differentiating software to ensure customers had many options and good price deals. In fierce aggressive mode against competitors, Travelocity and Expedia, Orbitz promised it would make good on its customer –care mission by being the first site with an advanced-technology search engine for a comprehensive listing of fares and schedules.

To be in the online travel service business, the real deal in staying alive, besides leaving no marketing opportunity unexplored, is to have a team that is very up on Web technologies. Considering Orbitz reached No.3 player-status in only 10 months of operation, you know the techies at Orbitz had to be doing something right. With 60 developers on its staff, would it be more accurate to say that the company is one big techie travel Web site? 

From day one, Orbitz avoided the big-iron Unix approach to provide information to site visitors and decided to go instead with hundreds of Red Hat Linux/Intel PCs for tasks such as searching for flights and fares for any-which-way-till-Sunday scenarios: I have a lot of money but no time; I have a lot of time but no money; I have to travel on February 3rd. For Orbitz, the secret lay in understanding the nature of the problem they were trying to solve.

Forget holistic approaches, no matter how complex the problem seems in packing up the entire family and booking a visit to Aunt Martha that includes a layover at the Mount Horub Mustard Museum. The problem is really just a discrete series of independent problems. As a result, there's no need for expensive big-iron SMP servers, nor is their a need for expensive high-performance compute clusters. Any of the simple discrete problems could be readily solved on a single PC server. The issue then was to load balance access to banks of PCs that were dedicated to solving particular classes of problems.

For the Orbitz planners, any dependence on big proprietary iron was met with outright derision. They talked about the advantage of doing tasks via distributed object computing, allowing themselves to scale different parts of the system independently to make focused changes such as scaling booking capabilities according to demand, without having to change everything in the system. Thus the theory went that they would be able to do the the tasks quickly and at less of a cost of a big-iron installation, which in turn would result in savings and efficiencies that they could pass on to both the suppliers and distributors in their value chain as well as consumers.

Today, Chicago-based Orbitz uses about 750 Linux-on-Intel computers in its data center. The return on investment is huge, according to Orbitz, in flexibility—adding or subtracting hardware as needed— and in cost savings, being able to accomplish more for less. Question is, in a business like this where technology innovation and marketing skills are required to be on par, how does a company like Orbitz keep both lights on? Who pushes decisions like Linux/Intel and how are they going to continue affordable innovations in the on-line travel market? Responding is Leon Chism, Orbitz Chief Internet Architect, one of the online travel industry’s key strategists eager to energize the on-line travel scene.

What does it mean to be ‘chief internet architect’ at a company like Orbitz?

Chism: Basically, my role is to shepherd the design and development of all the code that Orbitz uses, to ultimately deliver the end-user experience of the Web site. Included in my functions, too, is promoting solutions amongst both developers and senior managers, to get everyone to sign off.

What’s the balance of power like between you, for example, and the CTO?

Chism: The CTO has direct responsibility for hardware, software, networking, site operations. Most of my focus is on new development on the software side. As for how we work, we run a very ground-up organization. People in the trenches are making tactical decisions.

There are a number of tools that developers use here that folks like myself and other VPs really didn’t know about until they were running for some time, like CruiseControl.  So it isn’t a place where one person pontificates how we can reduce costs, but it comes from folks like our VP of technology development and from the director of network engineering.

Was it difficult making a case for Linux or other software that is Open Source?

Chism: It wasn’t a problem at all. The CTO was a Linux enthusiast and we always had production software running on Linux. Our Web servers are all Apache running on Linux. And a number of tools we use to assist in software development are Open Source. Some of our production systems are running Tomcat. All of us personally or otherwise used Linux in the past. You know, these days, anyone writing code for the Unix platform has done it on Linux on their own or at a prior job. Most kids coming out of college are using Linux. At this point, you easily find developers who understand the quality that you can get from this type of development environment—the challenge is convincing your boss.

Hardly the case at Orbitz?

Chism: The CTO was very much a fan of Open Source development. All of the VPs and directors had used lots of Open Source, and one real benefit is our Web site’s key component, the search engine, whose software runs on Linux.

Which Linux distribution do you use and do you depend on them much for support?

Chism: Red Hat. No, we don't. We are a very paddle-your-own-canoe operation. We hire capable people to support the code—all the code is written in Java— to maintain ourselves.

How many boxes do you actually have running Linux?

Chism: The number of machines is always expanding. It’s quite likely that by the time this goes to press, the number could be over 900 machines.

You more recently brought in 50 more Linux machines to replace Sun servers running Solaris. What are those 50 used for?

Chism: The earlier Linux PCs were all used for our low-fare search engine. The replacement machines that we added in September are for the Application Server layer, which runs the part of our application that specifically deals with Web interactions.

Are we talking about Orbitz running Linux clusters as in openMosix or Beowulf?

Chism: Our Linux servers are not clustered in the sense that I think you mean. We don't use Beowulf, or any other clustering software for creating one large "virtual server" out of many smaller machines. It never seemed appropriate to the problem we were trying to solve to introduce that kind of complexity into the system. We are able to generate a similar but not identical effect through the use of our own application-level load balancers and work coordinators as well as network load balancers. In essence, the "clients" of any one layer only know about a load balancer, and the load balancer is aware of all the machines that sit behind it and actually do the work. It forwards the request in such a way that it ensures the response goes to the correct client.

The clustering we do is basically for scalability. It allows us to add machines and to not be affected by individual failures—to add machines on the fly without any impact on the system’s client. Since day one, our goal has been to be the low-cost distribution channel. We knew that scalability and price-performance would both be critical elements to being a successful low-cost channel.

Might Orbitz be looking at Open Source alternatives to an Oracle database now or in the future?

Chism: When we first started we did look at some of the Open Source databases. We looked at MySQL seriously and the transactional security wasn’t quite up to snuff, but it is something we certainly keep our eye on.