In short, if the Expedias and Orbitzes of the world sell a ticket on American Airlines, Sabre computes the price for it. Our Low-Fare search is the
Our big differentiator is that our application checks availability in real time. So, when we display fares, we've actually verified that those seats are available at that price. Most of the other Web sites are just trying to predict [fares and availability} because they don't have the connectivity to the airlines that Sabre does.
Sabre has been a pioneer [with] our reservation system, and in the travel industry, in pushing the technology. Obviously, we've gone to a lot of trouble to make it work in real time.
What does Sabre ATSE do?
Basically, it's the architecture that describes the collection of hardware and the applications that run on it. There are various applications that run on ATSE: applications that keep up to date the availability, schedule, fares and rules for the entire airline industry; an application that prices tickets; and so on. Did you evaluate other operating systems?
The biggest driver was the IBM Transaction Processing Facility (TPF) [an online high-volume transaction processing platform]. We were running Hitachi mainframes on an IBM TPF platform. This [proprietary] code was written in templates when IBM mainframes were very expensive. But we were finding that the cost of mainframe technology in some areas was no longer appropriate. In the proof of concept, we set out to show that we could write totally portable code that would be C++ and could be run anywhere. We initially did that, but we knew we wanted to go to Linux long-term because we wanted an operating system that could run anywhere. People who knew Linux told us that there's a Linux compatibility mode, so that you could compile Linux programs on our existing proprietary Unix systems. There was really no value to going to a new proprietary Unix when all the Unix boxes are now having to support Linux anyway. What problems were you having with Low-Fare Lookup, and why did you think moving away from a mainframe-centric architecture could help solve them?
Low-fare searches are driven by Web customers and don't really not gain much revenue because the look-ratios are very high. A lot more people browse than really buy. So we needed to drive down the costs of the technology as much as possible. With that in mind, when we started looking at our reservation system in 2000, we knew that we wanted to go to Linux on commodity servers.
The low-fare search can be horizontally scaled, so with Linux we can just throw more boxes at it and, if one of those boxes fails, it really doesn't matter.
How did you get the project going?
We started with a proof of concept in 2000. We wanted the mainframe-like reliability for our core systems. So, we asked HP, which was then Compaq, to do a proof of concept on the HP NonStop architecture. It worked.
We developed a two-architecture strategy that calls for the HP NonStop core system transaction running data and the farm of Linux boxes doing the CPU- and memory-intensive work.
In late 2000, we made a presentation to senior management, got approval and went ahead with the project. Our team was composed of mainframe developers retrained in C++. We started in early 2001, and we had the first part of the system turned on in 2002. In parallel in 2002, we started doing the port, dealing with some small C++ and code issues, getting it to run on Linux and HP NonStop with identical code.
FEEDBACK: Is Linux destined to slay the mainframe?
Send your feedback to the SearchEnterpriseLinux.com news team.