Sonamine Website

  • Increase font size
  • Default font size
  • Decrease font size

Graph analysis scoring performance

AddThis Social Bookmark Button

In order to improve the Sonamine graph analysis platform between releases, we maintain two standard performance tests.  One is executed on a laptop and the other on a server cluster.  Here we describe the latest performance results of our laptop test using version 1.5 of the Sonamine Graph Scoring Engine.

Test data
The input data used for this test is anonymized data from a real life social network. There are 6.6 million edges and 1.9 million nodes.  This graph was imported into the SGSE as a csv file. The SGSE also scores each node depending on the characteristics of his neighbors.  In this test, we loaded up an additional file of 100,000 nodes which have certain criteria. The SGAS can therefore score each node's neighborhood based on these "matching" nodes.

Load time
The time taken to load the entire graph once into the SGSE was 45 min.

Performance
Eigenvector centrality (EVC) is useful in identifying the authority and influence rank of a node in the network.  EVC is an iterative function that stops when a certain precision (10E-9) is reached.  Generating EVC and exporting the results to a text file for all 1.9M nodes took 15 min 54 seconds.

We also generated the number of neighbors who were in the matching file.  This allows a telco to analyze each subscriber's immediate call circle with respect to many variables, including subscribers who churned and subscribers on another network.  It took 3 minutes 14 seconds to score 1.9M users.

We also generated the percentage of neighbors who were in the matching file.  This allows a telco to analyze each subscriber's immediate call circle with respect to many variables, including subscribers who churned and subscribers on another network.  It took 2 minutes 59 seconds to score 1.9M users.

Another test we run is to cluster the users together based on the density of links in each social group.  Grouping 1.9 M subscribers into their respective social groups took 7 minutes 23 seconds.

Hardware used
This performance test was conducted using a Dell M440 laptop with a quad core processor and 4GB of RAM.   It was running Windows XP.

 

User Login

               No account yet?