We might contribute some Avro benchmarks to this project:
http://code.google.com/p/thrift-protobuf-compare/
I suggest we contribute two benchmarks, one using the generic API and
one using code generation (specific).
To get an idea of what's benchmarked, look at, for example:
http://code.google.com/p/thrift-protobuf-compare/source/browse/trunk/tpc/src/serializers/ThriftSerializer.java
To me it looks like serialize() and deserialize() will be largely
impacted by setup, since the i/o streams, etc. are created per instance
serialized. And create() will be dominated by UTF-8 conversion. So
it's perhaps not the benchmark I'd write, but it is a benchmark.
Doug
