|
When Orbitz launched its travel service in June 2001, the company was already five years behind rivals Travelocity and Expedia. Despite the competition, the company knew it had an advantage because of its infrastructure.
March 12, 2002
URL:http://www.drdobbs.com/orbitz-reaches-new-heights/184414500
Most companies take precautions to avoid reinventing the wheel. But what if the wheel is in desperate need of reinvention? That was the problem that Chicago-based Orbitz faced as it prepared to enter the online travel industry in spring 2000. The travel-services provider had the option of either basing its system on a legacy application, or building an entirely new infrastructure. It chose the latter, and in less than six months became the third-largest and fastest-growing travel site in the U.S.
Orbitz CTO Alex Zoghlin describes the company's launch as a rebellion against Travelocity and Expedia, services that dominated the online travel market in 1999 and 2000. At the time of Orbitz's founding, Microsoft owned 70 percent of Expedia; and Sabre, a leading travel technology company, currently owns 70 percent of Travelocity. The two industry leaders had a five-year head start that included exclusive advertising and promotional deals with Web portals. (Travelocity powers travel sales on Yahoo and AOL, while Expedia sells tickets on Microsoft's MSN.)
Viewing Orbitz as a disadvantaged competitor may require you to suspend your disbelief however, because the company's founders are the five biggest United States airlines companies: American, Continental, Delta, Northwest, and United. This backing afforded Orbitz the money and time that it needed to challenge the other established services. Orbitz was prepared to compete on other levels, too. Says Zoghlin, "Taking on two giant corporations in a market they've dominated for five years—what could be more fun than that?" His theory was that if Orbitz wasn't completely different and better than its competitors from the start, it might as well not bother.
Zoghlin and his co-workers asked themselves how Orbitz could differentiate itself. The answer was with technology. The group figured that the average online travel customer in 2000 visited more than three Web sites before buying a ticket. No one site displayed all available flights, and consumers wondered whether there might be a different, better deal elsewhere. "I lovingly call this 'commerce interruptus,'" says Zoghlin. "You search, but you pull out at the last second."
This colorful metaphor set the stage for Orbitz's first and perhaps most difficult technical problem: providing more search results than competitors. And the company appears to have solved it. When I recently searched Orbitz for a round trip flight from New York to Los Angeles on an arbitrary date with no other restrictions, the results were 284 possible flights available for purchase directly from the site. Plus, there were hundreds more flights that required a phone call to the airlines for purchase. Travelocity, on the other hand, offered only eight available flights for the same journey. Expedia had just 31. (See the screenshot.)
It takes an airline-industry history lesson to understand why Travelocity and Expedia display fewer flights. Before Orbitz, most travel sites simply put a Web interface on 30-year-old technology. The computer reservation system (CRS) used in most U.S. travel bookings is dominated by old-school companies like Galileo, Amadeus, Sabre, and Worldspan. Orbitz worked with Worldspan during site development and uses Worldspan's technology as its booking engine, but saw too many drawbacks to using a similar technology for flight-searching functions.
If you've ever wondered how your travel agent finds such good deals, or been frustrated while experimenting with online travel sites, the inaccessibility of CRSs might have something to do it. For instance, each CRS system is different, and travel agents receive extensive training in how to properly use the command-line interface and extract the right information from it. People other than travel agents can access CRS through a Web site, but few know a travel agent's tricks of the trade.
For Orbitz, it didn't help that CRSs rely on mainframe computers (Worldspan and Sabre use IBMs, Amadeus uses a Unisys ES7000 Intel-based mainframe, and Galileo uses an SNA-based mainframe). As Orbitz's press materials put it, "using a mainframe to keep up with the dynamic needs of online travel searching is like trying to turn the Titanic around in a bathtub."
A standard search for a trip from New York to L.A. could have two billion possibilities, because a flight could theoretically route through every airport in the country. To minimize transaction time, mainframe operators have optimized the algorithm to search several hundred thousand and a million possibilities. If a mainframe tried to search all two billion combinations, says Orbitz CIO Kevin Malover, "the system would be brought to its knees." An extensive search engine would take computing capacity away from other tasks, like flight booking capability.
Zoghlin is quick to note that he's "not a mainframe basher," describing them as "amazing" and "revolutionary" for their time. But he refers to Moore's Law, which states that the number of transistors capable of being squeezed onto a chip (roughly equivalent to computing speed) doubles every year. With technology improving so quickly, why use systems built decades ago, even if they were good systems in their time?
To solve the mainframe problem, Orbitz brought in ITA Software, a Cambridge, MA-based travel technology company founded by computer scientists from MIT's Artificial Intelligence Laboratory. Founded in 1996, ITA built its reputation by licensing software to big names in travel technology like Amadeus, Logibro, and SITA.
In 1999, after seeing ITA demonstrate its system at a conference, representatives from the five major U.S. airlines approached ITA about Orbitz. ITA CEO Jeremy Wertheimer, notes that the meeting took place "before [Orbitz] had any employees or anything." Yet, by spring 2000, ITA had signed an agreement to provide the flight search engine for Orbitz.
The group decided to build Orbitz Flight Search Engine with the help of distributed object computing. This technique divides client/server applications into components that can function across different networks and operating systems—nearly the opposite of a mainframe. The benefit is a more flexible, scalable IT infrastructure. For instance, Orbitz has 600 PCs, each running a different part of the system software. If the company finds a problem in one part of the engine, administrators can swap in new PCs and software without having to overhaul everything.
The PCs are Intel boxes that mostly run Linux. Wertheimer says ITA is dogmatic about having Linux as its operating system. The group also feels strongly about adhering
to Intel hardware because of its sound performance.
"The actual computation that gets performed at Orbitz is tremendously expensive," explains Wertheimer, "It would be sort of unthinkable on a mainframe." For instance, computer interactions back and forth between the airlines are complicated, as schedules and fares change frequently. Orbitz draws that information from a data center in Atlanta
that broadcasts it in real time.
Other travel companies appear to be following Orbitz's lead. In late 2001, Sabre announced that it would move its airline-registration database system from mainframes to Compaq servers, a change that will take Sabre several years and $100 million to implement.
The software running on the network of PCs was the result of work that Wertheimer started at MIT in 1992. It's the same software that all of ITA's clients use, but the company had never had a client as big as Orbitz. Fortunately, says Wertheimer, more than 90 percent of what Orbitz needed, aside from routine bug fixes and software updates, had already been completed. ITA never needed to send more than a half-dozen employees to Orbitz.
ITA engineers wrote the entire flight search engine package themselves—there aren't any commercial databases here. (Although, outside the flight search engine, Orbitz does use Oracle 8i and other third-party applications.) The homebrewed system lets Orbitz increase the site's functionality on the company's own schedule. While competitors must wait for an upgrade to third-party software before performing quarterly or twice-yearly updates, says Zoghlin, Orbitz refreshes its code two or three times a week, adding features, fixing bugs, and optimizing algorithms.
Because of ITA's background in computer science and artificial intelligence, the founders weren't afraid to consider languages outside the normal realm of travel technology. For instance, the conventional programming language for building travel software is Assembly—yes, the same Assembly language from the early days of computing that's just one step up from machine code. Because it was building a Web-based system from the start, Orbitz ditched the travel agencies' assembler legacy and went with Java for about 80 percent of its development efforts. Code that touches data is written in C++, which is considered a more compact and efficient language for performing data-handling tasks.
The high-level algorithms are almost entirely in Lisp, one of the oldest programming languages. You're excused for chuckling, or saying "Why Lisp?" Although the language can be inefficient if used without extreme caution, it has a reputation for compactness. One line of Lisp can replace 20 lines of C. ITA's programmers, who learned the language inside and out while at MIT, note that LISP is highly effective if you ditch the prefabricated data structures. "We're something of a poster child for LISP these days," says Wertheimer. "Lisp vendors love us."
The downside of using a sometimes-maligned language is that it's hard to find good Lisp programmers. Today, only half of ITA's coders are Lisp gurus. For its own Web site, ITA relies on server-side Java, partly because of the availability of capable Java programmers.
In addition to being Lisp fans, ITA's employees have a more modern love. Wertheimer describes himself and his colleagues as "serious XML bigots," using the language to communicate with airlines about flight schedules, delays, and gate numbers.
XML is sometimes accused of being inefficient, but ITA thinks the benefits are worth it. When designing XML documents,
ITA programmers were able to choose
human-readable names for data, like CityorAirportCode (see Example 1).
And there's no arguing with the fact that XML beats the travel industry's old method of displaying data: screen scraping. Screen scraping is used to read data formatted for an older terminal and to reformat it for, in this case, a Web browser. In the reverse direction, user input on the new interface must be reformatted for use by the old device.
example 1: Orbitz's search engine queries airline schedules by
sending XML tags like these within a larger document.
With screen scraping, the formatting is thrown off every time a byte of information changes on the original screen. With XML, the data is transferred separately from the screen design. "When byte six moves to byte eight, you don't have to change your code," explains Wertheimer. "None of your code ever has to change, even as we add feature 4000."
On the front end, the interface for Orbitz Flight Search Engine offers users an easy way to compare flights. This is another way in which Orbitz differs from other travel services. On most airline-ticketing sites, users must scroll down a Web page to view available flight information. With Orbitz, search results are shown in a grid at the top of the page. From there, customers can narrow the search by sorting flights by airline, price, or number of stops.
The interface was the result of extensive usability research. Zoghlin and Malover believe strongly in the importance of human factors and cognitive sciences. They hired experts in these fields and had them design Orbitz's interface with the help of quantitative research. The company performed usability tests with people of various demographics in Orbitz's target market. The scientists and testers both evaluated the site's use of language, text, graphics, and color, and the company refined the interface based on feedback. Customer response has been positive, and Orbitz is in the process of applying the grid display to its car rental and hotel reservation searches.
Orbitz launched its public site in June 2001. From concept to completion, it took the company merely eight months to build the infrastructure and implement the software. Orbitz must be keeping its customers happy, because site traffic is high. According to online-audience measurement firm Nielsen/NetRatings, 6.3 million unique users visited Orbitz.com in October 2001. In November 2001, Orbitz was selling 30,000 to 60,000 tickets per day. Malover says the site sees 750,000 to one million airfare searches per day.
To ensure that the site stays online even in the midst of high demand, Orbitz worked with IBM to develop its traffic management configuration. Orbitz now uses load management tools from Mercury Interactive. Malover notes that the distributed computing system lets the company increase server capacity as needed. Adding new PCs to the network is a quick and inexpensive task.
Today, Orbitz remains focused on maintaining its competitive advantage through technology. Of about 160 employees, nearly 100 work on the software and hardware for the site. In the coming year, Malover says they'll focus on platform decisions, including the transition to more Linux servers.
Company executives seem satisfied with Orbitz's progress so far, but are intent on continuing to innovate. When asked what he learned from the development process, Zoghlin replies, "Don't be afraid to question conventional wisdom." Convention certainly wasn't an issue for Orbitz, and neither was reinventing a wheel or two along the way.
Jen is a journalist based in Hamburg, Germany. She has written for The Industry Standard, Publish, Lycos, and several other publications. You can reach her at [email protected].
Terms of Service | Privacy Statement | Copyright © 2024 UBM Tech, All rights reserved.
The Legacy
Beyond Mainframes
Disassembling
<CityOrAirportCode>BOS</CityOrAirportCode>
</Origins>
<Destinations>
<CityOrAirportCode>LAX</CityOrAirportCode>
</Destinations>
<DepartureTimeRange>
<EarliestTime>2002-1-15T05:00</EarliestTime>
<LatestTime>2002-1-15T23:59</LatestTime>
</DepartureTimeRange>
Carrying the Weight
