Hadoop tried Avro for its RPC, but there were too many challenges. We backed out to ProtoBuf, which has worked well for us. In ORC, I used ProtoBuf for the metadata (and not the data) and it has also worked well for that.
-- Owen On Wed, May 8, 2013 at 8:42 AM, Henry Saputra <[email protected]>wrote: > Hi Hyunsik, thanks for revisiting this Avro switch discussion. I agree to > punt this discussion into the future. > Even now Hadoop uses protocol buffer for its RPC. > > I am ok to not open JIRA for it. I think once we have one or two good > release then we could revisit this discussion again. > > Really appreciate for opening up discussion about this. > > > - Henry > > > On Wed, May 8, 2013 at 4:35 AM, Hyunsik Choi <[email protected]> wrote: > > > Dear Henry and folks, > > > > In TAJO-28 (https://issues.apache.org/jira/browse/TAJO-28), there was a > > discussion of switching protocol buffers to another one. Henry's > suggestion > > is to switch protocol buffer to avro. At that time, I said that I'll > create > > jira issues for it. I'm sorry for late response. > > > > At the first time, I also +1ed for the switching. However, Tajo has > heavily > > used protocol buffers throughout the project. It will cause many > overheads > > for us. If you are ok, I would like to postpone this discussion. I think > > that there are many important issues to be solved right now. > > > > Best regards, > > Hyunsik > > >
