Alex JF

Optimizing Synchronous Online Computation of Large Graphs

  • Category: Thesis
  • Authors: Alexandre Fonseca
  • Start date: 2014/01
  • Status: Finished

This was my master thesis project developed at Telefónica between January and July 2014 with academic supervision from Universitat Politècnica de Catalunya.

In this thesis, I try to exploit the idle times inherent to imbalances in the BSP/Pregel execution model (used by Giraph) to speedup the computation of large graphs. This is done in the context of a system being developed at Telefónica called Realtime Giraph that enhances Giraph by providing a realtime incremental computation mode.

Realtime Giraph is part of the project.

The 3 mechanisms implemented to achieve the objective were:

  • Pipelining of simultaneous event execution.
  • Partition-level local synchronization.
  • Vertex-level local synchronization.
comments powered by Disqus