Union Pacific Railroad Moving from Mainframes to Blades

Union Pacific, which almost 40 years ago was the first railroad to schedule car-level shipments thanks to its IBM mainframe, is in the process of moving to a distributed IT infrastructure in which smaller blade servers from Dell and HP eventually will replace the mainframe. Railroad officials say that new business demands, the growing use of open-source technology and the aging mainframe programmer population are among the key reasons for making the move, which is costing Union Pacific more than $150 million and is scheduled to be complete by 2014.

In the 1960s, Union Pacific became the first railroad in the world to schedule each shipment at the car level and automate the tracking of the cars when it brought in an IBM mainframe.

Forty years later, the railroad is in the midst of a multiyear project that will see a new distributed network, comprising thousands of x86 servers-primarily blades-running Linux, replace the mainframe.

After four decades and 11 million lines of macro assembler code, it was time to rethink how the railroad was operating, Lynden Tennison, senior vice president and CIO at Union Pacific, said in an interview.

"It's been very effective running the railroad for a long time," Tennison said of the IBM mainframe. "But things change."

New business demands, the move to open-source technology, years of acquiring and absorbing other companies and their IT infrastructures, and the aging of the mainframe programmer population are among the key drivers for the change, he said.

What are the advantages of mainframes? Or of x86 systems? Click here to find out.

At the end of the project, which will cost the railroad between $150 million and $200 million, Union Pacific-with its 8,400 locomotives and 32,000 miles of tracks that run through 23 states-will have an IT platform that is event driven, built via SOA (service-oriented architecture) and running Linux as the primary OS. Tennison said he hopes to shut down the mainframe in 2014.

There won't be any big event signaling the end of the mainframe's life, he said. Rather than make a massive shift over a weekend from the mainframe to the distributed platform, Union Pacific instead will spend the next five years adding and turning on new x86 systems so by the time the mainframe is turned off, the new, completed infrastructure already will be up and running.

The platform is being built in two data centers in Omaha, Neb., on blade systems primarily from Dell, though the railroad does have some systems from Hewlett-Packard.

"We intentionally want to keep it that way so that [the servers are not from] a single vendor," he said. "We wanted to be a commodity-based infrastructure."

The systems are running Red Hat Linux, with middleware from Oracle, Tibco Software and BEA Systems. Union Pacific had already been using applications from these vendors. Tennison said the middleware already worked well, and that given the size of the project, it was better not to introduce products from new vendors if they weren't needed.

Union Pacific isn't using much server virtualization, Tennison said. The IT platform is being built on blade servers that cost about $3,000, so it isn't a challenge financially to add more servers as needed. The railroad also saw some performance problems several years ago when it started using VMware virtualization on large, eight-socket systems that cost $100,000 or more.

Now Union Pacific is focusing on smaller servers, many of them running Opteron processors from Advanced Micro Devices, though the railroad will use Intel-powered systems as well. Currently, most of the chips are quad-cores, though there are plans to begin using AMD's six-core "Istanbul" Opterons.

In addition, whatever virtualization is being used is coming from the operating systems themselves, in particular Red Hat's Linux virtualization capability and Microsoft's Windows Server 2008 Hyper-V for Windows.

A key part of the infrastructure is a transportation system called NetControl, which will replace the mainframe-based Train Control System. Tasks such as order taking, monitoring and scheduling shipments and train schedules, and dealing with service interruptions will fall to NetControl.

NetControl is about a third complete, and already is handling some transaction duties, such as bill-of-lading. The new platform is more forgiving of such errors as misspellings than the mainframe, which is making the process faster and more efficient. Eventually Union Pacific will tie NetControl into a nationwide automated system-called Positive Train Control-which is designed to increase track safety and reduce collisions. The federal government wants that system in place by 2015.

Union Pacific also has switched to a new SAP ERP (enterprise resources planning) system that is now running on the distributed system.