Description
Hadoop could benefit greatly from a simple record I/O framework that enables the specification of simple record types and enables the generation of code for serialization/deserialization in multiple target languages. The framework would handle a small well understood set of primitive types and simple compositions of these (structs, vectors, maps) . It would be possible to leverage this framework to express I/O in MapReduce computations and to use this as the basis for Hadoops RPC implementation. This would make interfacing with code in languages other than Java much easier.