Currently Hadoop supports only three basic integer-like types: ByteWritable, IntWritable and LongWritable. They provide a fixed tradeoff between their value range and on-disk space consumption. But it is sometimes useful to be able to store integer values with broader allowed range, but less space consumption when possible.
This is especially useful when storing very long series of values, combined with delta encoding.
Lucene already implements variable-length encoding for positive int and long. I propose to add similar Writable implementations, which use the same encoding methods.