Scalability and Performance of Multi-threaded Algorithms for International Fare Construction on High-Performance Machines
We describe the design, implementation and performance of a project for constructing international fares at United Airlines. An efficient fare construction engine allows an airline to simplify, and automate the process of pricing fares in lucrative international markets. The impact of a fare engine to the revenues of a large airline has been estimated at between $20M and $60M per year. The goal is to design an efficient software system for handling 10 Gb of data pertaining to base fares from all airlines, and to generate over 250 million memory-resident records (fares). The software architecture uses a 64-bit, object -oriented, and multi-threaded approach and the hardware platforms used for benchmarking include a 24-CPU, IBM S-80 and 32-CPU, Hewlett-Packard Superdome. Two competing software designs using (i) dynamic memory and (ii) static memory are compared. Critical part of the design includes a scheduler that uses a heuristic for load balancing which is provably within a constant factor of optimality. Performance results are presented and discussed.
KeywordsLink Formation Dynamic Memory Fare Class Compatibility Check Destination City
Unable to display preview. Download preview PDF.