Applications

Tangram is flexible and can express a wide variety of workloads efficiently. Here, we provide a subset of applications that can be supported by Tangram as examples. In fact, Tangram can support more different applications.

• Bulk Processing

  As MapUpdate is a superset of MapReduce, it should be able to support any MapReduce-style bulk processing applications. We provide two simple examples:

  • Word Count
  • TF-IDF

• Machine Learning

  Tangram supports parameter-server-style distributed machine learning naturally with the MapUpdate API. Parameter-server is the de facto solutions for distributed machine learning. Please refer to the PS and Bosen papers for more information. Tangram also supports asynchronous training with staleness.

  • Logistic Regression with SGD
  • K-means

• Graph Analytics

  Tangram also naturally supports vertex-centric graph analytics.

  • PageRank
  • SSSP

• Other Asynchronous Workloads

  • Distributed Crawler
  • NOMAD
  • Graph Matching