StarTrack next generation: a scalable infrastructure for track-based applications

  • Authors:

  • Maya Haridasan;Iqbal Mohomed;Doug Terry;Chandramohan A. Thekkath;Li Zhang

  • Affiliations:

  • Microsoft Research Silicon Valley;Microsoft Research Silicon Valley;Microsoft Research Silicon Valley;Microsoft Research Silicon Valley;Microsoft Research Silicon Valley

  • Venue:
  • OSDI'10 Proceedings of the 9th USENIX conference on Operating systems design and implementation
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

StarTrack was the first service designed to manage tracks of GPS location coordinates obtained from mobile devices and to facilitate the construction of track-based applications. Our early attempts to build practical applications on StarTrack revealed substantial efficiency and scalability problems, including frequent client-server roundtrips, unnecessary data transfers, costly similarity comparisons involving thousands of tracks, and poor fault-tolerance. To remedy these limitations, we revised the overall system architecture, API, and implementation. The API was extended to operate on collections of tracks rather than individual tracks, delay query execution, and permit caching of query results. New data structures, namely track trees, were introduced to speed the common operation of searching for similar tracks. Map matching algorithms were adopted to convert each track into a more compact and canonical sequence of road segments. And the underlying track database was partitioned and replicated among multiple servers. Altogether, these changes not only simplified the construction of track-based applications, which we confirmed by building applications using our new API, but also resulted in considerable performance gains. Measurements of similarity queries, for example, show two to three orders of magnitude improvement in query times.