Hi, In MapReduce system, map function takes one <K, V> pair as input (sequential read). BTW, Google Pregel's compute() function takes argument, 'MessageIterator* msg'. In other words, compute() function receives messages from the start.
It's my very rough idea but, I think, we can broadcast initial messages to each tasks (or BSPPeers) flexibly. https://issues.apache.org/jira/browse/HAMA-258?focusedCommentId=13018842#comment-13018842 I'm not sure that we really need to implement the input/output system. Any comments are welcome. Thanks. -- Best Regards, Edward J. Yoon http://blog.udanax.org http://twitter.com/eddieyoon
