Test performance of vector computation

[yunseong]

This is part of #265

@gyeongin and @beomyeol reported that Vector computation in Dolphin was too slow.

First thing we can try is replacing vector library rather than using Mahout's. FYI. Vortex used a vector that was implemented using array and map (Two implementations exist which I need to cleanup). You can see the code (https://github.com/yunseong/reef/commits/cleanup-with-hdfs).

I took another look at Spark MLLib shortly, and they seem to use breeze (https://spark.apache.org/docs/1.1.0/mllib-guide.html) for linear algebra library. Sorry for confusion. We may also want to check this.