Gumball: a race condition prevention technique for cache augmented SQL database management systems

  • Authors:
  • Shahram Ghandeharizadeh;Jason Yap

  • Affiliations:
  • University of Southern California, Los Angeles, California;University of Southern California, Los Angeles, California

  • Venue:
  • DBSocial '12 Proceedings of the 2nd ACM SIGMOD Workshop on Databases and Social Networks
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Query intensive applications augment a Relational Database Management System (RDBMS) with a middle-tier cache to enhance performance. An example is memcached in use by very large well known sites such as Facebook. In the presence of updates to the normalized tables of the RDBMS, invalidation based consistency techniques delete the impacted key-value pairs residing in the cache. A subsequent reference for these key-value pairs observes a cache miss, recomputes the new values from the RDBMS, and inserts the new key-value pairs in the cache. These techniques suffer from race conditions that result in cache states that produce stale data. The Gumball Technique (GT) addresses this limitation by preventing race conditions. Experimental results show GT enhances the accuracy of an application hundreds of folds and, in some cases, may reduce system performance slightly.