Company runs mainframe applications on cheaper swath of x86 servers.
For UPS, is not about how to get more horsepower for demanding workloads; it’s about consolidating, streamlining and using technology to get an edge on the competition.
“Using technology to differentiate ourselves from our competitors has always been fundamental to our success, and it’s one of the reasons we’re moving forward with [grid] technology,” Brian Cucci, manager of the Advanced Technology Group at UPS said during a Webcast last week with DataSynapse, the company that supplies the Atlanta-based company its grid software.
The software, called Grid Server, is now in production use at UPS and lets the company distribute a billing invoice application that once ran on an expensive mainframe across a group of cheaper x86 systems running Linux.
“Most of the organizations we spoke to adopted grid in a little different way. They pursued it to solve that killer business problem such as not having enough horsepower with their existing infrastructure,” Cucci said.
For UPS, grid computing was just another piece in its evolving IT puzzle, which is aimed at reducing costs and improving efficiency. The company’s Technology Directions Subcommittee, which is made up of representatives from across the organization and reports to the CIO, is charged with keeping track of hot technologies, determining which can best bring business value.
Grid computing gained priority and moved to the top of the group’s radar screen last year, because it fit in nicely with several other technology projects that either were underway at UPS or were in the planning stages, Cucci said. Those projects include virtualization and consolidation efforts, as well as an initiative to move to a computing-on-demand approach to IT that focuses on the use of low-priced commodity hardware.
“We’re moving away from buying specialized servers to host one application on,” Cucci said. “Over the past decade, we were in that mode, and we’re trying to get to a more commodity, capacity-on-demand approach.”
DataSynapse’s GridServer helps move UPS in that direction. The software pools commodity hardware into a grid, or virtual pool, in which compute resources are allocated on-demand and on-the-fly according to application needs. The software is policy based, meaning customers can set priorities so that applications get the horsepower they need when they need it in order to meet service level agreements.
Cucci says the maturation of Linux into a stable operating system paired with the increasing performance of x86 hardware enabled UPS to think seriously about moving workloads off of mainframes. Windows also plays into the equation, he says. “We’re basically reviewing systems on a case-by-case basis to select the best platform for them,” he says.
UPS started researching grids in 2005. “Grid has been around for a long time, but we felt it was time that it was enterprise ready to run internally within the walls of our company,” Cucci said.
After looking at a number of grid-software vendors, which Cucci did not name, UPS settled on DataSynapse because it had a handful of customers that were using the software in a way similar to how UPS wanted to use grid technology, Cucci said.
UPS did a lab proof of concept with DataSynapse GridServer in 2005 and found that an analytics application that had taken 9.5 hours to process in its traditional environment was completing its tasks in 45 minutes on the grid.
Convinced by the success of the initial trial, UPS installed grid infrastructure in its two data centers and began moving the billing invoice application off one of its 15 mainframes and onto a grid of about a dozen commodity servers.
“We didn’t do the big bang approach, taking the whole project and throwing it on the grid at once,” he says. “We did a cutover method, cutting over portions weekly.”
UPS completed the migration late last month.
The billing invoice composition application is critical for UPS’ business, because it produces invoices and billing statements – between four and six million pages weekly – for customers. The program uses a large amount of mainframe processing power, which has meant spiraling software licensing costs as the number of mainframe engines needed to run the application grew.
“By moving [the compute intensive part of the application] to Linux we were able to contain these costs,” Cucci said, though he didn’t place a dollar value on the savings.
In addition, UPS has reduced the amount of time it takes to process the invoices, he said.
Bringing grid technology into an organization isn’t always easy and having the overarching technology subcommittee supporting the software was a definite plus for UPS, said Mike Smith, lead systems architect in the UPS Advanced Technology Group.
“By presenting [grid] as a solution to business problems, I think our business users could easily see that grid wasn’t your conventional scientific computing in the supercomputer data-center role,” he says. “They could see results as far as increased performance with their applications.”
