+1 for switching to Avro.
Actually, I have considered the switch from protocol buffer to Avro. In addition, tajo-rpc could be replaced by avro-rpc. If we switch tajo-rpc, we can concentrate on fewer works. It may be the right time to do them because Tajo is under early stage and is an incubator project.
It will be very tedious work because many (de)serializable objects are implemented by protocol buffers. It would be good to create many subtasks for them. If there are not any objection until this weekend, I'll an umbrella jira issue for subtasks. It also would be a very good opportunity to create issues for newbies.