The MPP stands for massively parallel processing, an approach to looking up information that is used by Google and other search engines. It is based on multiple databases on multiple servers, each holding a portion of the total database. A query is subdivided to execute on each of the appropriate servers and then the results combined, speeding up the systems ability to come up with answers to complex queries from masses of data. Another example is the Navy parallel processing system Kraken.
Frontier CIO Robert Rapp says the airline's yield management process runs on Bizgres MPP. The system predicts the yield or profit that Frontier will receive on various flight combinations and ticket prices. The system helps Frontier determine where to offer seats at bargain prices and where to avoid what might turn out to be a competitive bloodletting, with no one profiting, says Rapp, the former CIO of Southwest Airlines, a pioneer of low-priced flights.
"Greenplum allowed us a very economical solution for a mid-sized airline. There are large amounts of parallelism in the system," says Rapp. A comparable but higher end commercial system used by retailers such as Wal-Mart comes from Teradata, a unit of NCR Corp. "Greenplum was available at 20-30 times less" than such a system."It was available at a very nice price point for us," adds Rapp.
Frontier used the system for 5-6 months before actually putting it into production in February. The 2.0 version of Bizgres MPP came out at the end of 2005. The actual coding of the yield management system was done by a private business analytics consulting firm, Daxpy LLC, in partnership with Greenplum, which will be offering the yield management system as an application that can be added onto its Bizgres MPP tools and the PostgreSQL database.
The system runs on 4 Dell 2850 servers, including a controller server and three database servers, says Dave Croiter, leader of the Frontier project. The system can be upgraded by adding more Dell hardware or new dual-core hardware. Ticket purchasing information is constantly available to the system in an updated fashion. A query that must access 800 million records will return a result "in a few seconds or, at most, 15 seconds, not 20 minutes," giving business analysts at Frontier the ability to respond rapidly to competitors' offerings, Croiter says. Various alerts are built into the system to spot regions where sales are not living up to predicted levels so business analysts can wade into the data to try to determine why. "In the past, we discovered the why when it was already too late and the revenue was lost," he says.
Sometimes airlines sell distressed inventory when they have too many seats and too few passengers. Yet, sometimes it doesn't make sense to match cuts made by a competitor and lose money. "If a competitor takes drastic action, an analyst knows whether to add matching fares, add flights to Frontier's schedule, or do nothing." he said.