Right now we have a Hadoop job in Java that is working with some C++ binaries. We write files to NFS and C++ and Java read them and that is our form of communication, which prevents us from scaling. I'm looking into Proto Buff, Thrift and Avro to get away from the NFS thing. This approach would definitely be better than the NFS approach, right?
Asked
Active
Viewed 710 times
1
-
Why not using HDFS and SequenceFiles? – Thomas Jungblut Apr 21 '11 at 17:26
-
That would mean my C++ would also have to interact with HDFS? I think that will be a little hard. I am thinking about JNI also, but unsure about it. Do you think its not a very good idea to use these RPC frameworks? – Meg Apr 21 '11 at 17:45
-
Your C++ can interact with HDFS by FUSE http://wiki.apache.org/hadoop/MountableHDFS – xboard Apr 21 '11 at 21:43