Hi, Currently, we're using Hadoop RPC to communicate and transfer messages between BSP workers. Later, we may want to use Thrift[1] or Protobuf[2] for better performance. As far as I know, Protobuf and Thrift are faster than Hadoop RPC.
What do you think? 1. http://en.wikipedia.org/wiki/Apache_Thrift 2. http://code.google.com/p/protobuf/ -- Best Regards, Edward J. Yoon http://blog.udanax.org http://twitter.com/eddieyoon
