On Thu, May 20, 2010 at 09:29, Erik Forsberg <forsb...@opera.com> wrote: > Assuming we're building a web interface that needs to read some files > from HDFS, and we don't want to use the Java API, would the Thrift > gateway be the best option? Or is fuse-dfs better?
I wrote namenode and datanode plugins to expose HDFS through Thrift long ago. I then moved to fuse-hdfs because of performance issues (reading data through Thrift was slower than through libhdfs, which is what dfs-fuse does). > How about using the > http APIs for listing and fetching data, how will they compare > performance-wise? I remember having done some tests, and reads through the HTTP API were also slower than libhdfs reads. C
