Spark: matcher response time too long as barefoot servers(stand-alone)

[ZhiWeiCui]

I follow this suggestion: #59

For Spark, we upload the bfmap file to HDFS. This allows the Spark executors read the map from HDFS, which should be better than having them all connecting to the PostgreSQL database at the same time.

To create a bfmap file from the PostgreSQL database, use this code: https://gist.github.com/jongiddy/67c7ace4e7394e1e5f3bea978ddf74ec (this is set to run inside a Vagrant virtual machine, but changing the hardwired /vagrant paths will make it suitable for other environments).

To read the bfmap file from HDFS, we created a HadoopMapReader: https://gist.github.com/jongiddy/b68be517274a424df84d2bea4cdd6354

And our BroadcastMatcher then looks like this (although I have edited out some application-specific code): https://gist.github.com/jongiddy/286857e09f9881854a725634ca82b515

I can get the correct results, but it takes a long time as Barefoot servers(stand-alone).