[SPARK-4129] Performance tuning in MultivariateOnlineSummarizer

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Component/s: MLlib
• Assignee: DB Tsai
• Reporter: DB Tsai
• Created: 29/Oct/14 05:39
• Updated: 29/Oct/14 17:15
• Resolved: 29/Oct/14 17:15

Description

In MultivariateOnlineSummarizer, breeze's activeIterator is used to loop through the nonZero elements in the vector. However, activeIterator doesn't perform well due to lots of overhead. In this PR, native while loop is used for both DenseVector and SparseVector.

The benchmark result with 20 executors using mnist8m dataset:

Before:
DenseVector: 48.2 seconds
SparseVector: 16.3 seconds

After:
DenseVector: 17.8 seconds
SparseVector: 11.2 seconds

Since MultivariateOnlineSummarizer is used in several places, the overall performance gain in mllib library will be significant with this PR.

Issue Links

• links to: [Github] Pull Request #2992 (dbtsai)