When Cendant Corp.'s Travel Distribution Services (TDS) division considered shifting its airline-fare system to Linux on Intel-based servers, the IT department couldn't simply flip the switch. The system handles 700 transactions per second in the course of processing millions of fares from more than 500 airlines around the world. So the IT team set aside a few months to do a careful analysis of the business case.
Now that Linux is more commonly viewed as a mainstream option for mission-critical functions, IT managers are increasingly evaluating the open-source operating system with the same due diligence with which they compare commercial offerings, according to industry analysts.
"It's not a science project anymore," says Julie Giera, an analyst at Forrester Research Inc. "At this point, Linux shouldn't be different than any other commercial software package you buy. The rules should be the same. The level of scrutiny should be the same, and the process for approval should be the same."
A key first step is establishing the criteria by which Linux will be judged. Enterprise Linux use has concentrated on the server, and decisions are generally made in concert with moves to cheaper hardware. So the business case is usually built for the hardware and the software operating system at the same time.
Two years ago, Chicago Mercantile Exchange Inc. (CME) was paying $20,000 to $40,000 for each of its Sun Microsystems Inc. servers running Solaris, according to Joe Panfil, the company's director of enterprise technology. When the CME needed to add capacity, the IT team was anxious to see if it could reduce the server costs.
Linux servers were priced at about $3,000 apiece, and a Red Hat Inc. support subscription tacked on another $400 per box, Panfil says. Even though the operating system can be downloaded for free, serious users typically don't want to risk running Linux without a support contract, especially if they're running mission-critical applications on it.
But cost wasn't the sole metric in the business case, particularly after Sun began to drop its server prices. The CME had to be sure that its critical third-party software products—Tibco Software Inc.'s middleware, BEA Systems Inc.'s WebLogic application server and Oracle databases—were certified to run on Linux, Panfil says.
Transaction speed was another key driver. The CME makes money based on the number of trades it can process, so every millisecond it shaves off the round-trip trading time counts.
But none of that would matter if the system didn't run reliably on Linux. Internally developed electronic trading, clearing and regulatory applications needed to be ported to Linux, and developers needed training to write code optimized for Linux.
Proving the Case
Once the metrics are established, it's time to test. For Orbitz Inc., that meant bringing together four members of its software team and four members of its hardware engineering team when the leases for the Sun servers that ran its BEA application servers were due to expire in the summer of 2002.
On paper, Linux made sense for Orbitz. The Chicago-based online travel service had the skills, infrastructure and tools to work with the open-source operating system, since the low-fare search engine it licensed from ITA Software Inc. already ran on Red Hat's Linux distribution.
But Orbitz still needed to make sure its WebLogic application servers would perform as well on Linux on Intel-based hardware as they had on Sun Solaris servers, taking into account new functionality the travel company was planning for its site.
So the Orbitz IT team consulted with the finance and product marketing departments to find out which new features they wanted and how much additional traffic they expected. Orbitz architects estimated what it would take to deliver the new features, and then systems engineers determined the hardware capacity that would be needed.
Orbitz did a CPU-for-CPU comparison of Intel Corp. processors running Linux against Sun Sparc processors running Solaris and found that the Intel CPUs performed twice as well, according to chief Internet architect Leon Chism. Orbitz then calculated the incremental cost of purchasing new servers from Sun and compared that with the amount it would spend if it adopted the open-source model and used greater numbers of smaller commodity servers. It also factored in the additional overhead required to manage the Linux servers. "We did that business case" over three months, says Pete Stoneberg, director of systems engineering, "and it clearly came out in the open-source Linux camp."
Cendant TDS built a lab to test 3-GHz Intel chips on eight-way IBM servers against the 900-MHz Sparc chips it had been using on 24-way Sun boxes. The goal was to see if its 360 Degrees Fares application could scale out through smaller, redundant Linux servers and reliably process an equivalent number of transactions in the same amount of time as the larger, more expensive Unix hardware.
"For our company, stability is important. We believed we could get high levels of stability through a highly redundant system built on lots of low-cost, high-performing Intel boxes," says Robert Wiseman, chief technology officer at Cendant TDS. "It turned out, for this application, we could run at least as many transactions through the Intel boxes as the Unix boxes."
Final Tweaks
That wasn't the end of it. The team ran the application for 30 days and found Unix more forgiving of problems such as memory leaks. Developers spent about three months tweaking the application code to deal with the slight operating system differences between Unix and Linux. "But at the end of the day," Wiseman says, "the redundant architecture we created with the Lintel environment gave us better stability."
The next step was determining the number of boxes needed, based on the number of transactions the hardware is capable of handling, and determining the headroom Cendant TDS wanted above the peak load. The numbers told the story.
"The cost of building out our platform on Lintel versus continuing to build on Unix was 90% less expensive," says Wiseman. "It was dramatic."
The business case for Linux also won the day at the CME, saving the exchange an estimated $2.8 million last year. "We had a lot of preplanning and thought in front of the move," Panfil says. "Where we've needed faster servers and cost reductions, we've implemented it, and we're happy."
But the evaluation is ongoing. "We're always going to be looking at new technologies," he says. "If Solaris 10 proves to be just as fast as Linux and more reliable, we'll implement it on commodity servers."
| Open-Source Costs, Benefits and Risks | |
|
