As discussed on hadoop-general, we need somewhere to host the non-HDFS filesystem clients. S3/S3N, ftp are all in hadoop-common, with the JAR dependencies there. This doesn't scale to openstack, azure, or handle changes in the S3 dependencies.
With a project of hadoop-common/hadoop-filesystem-clients, we could add separate FS clients: hadoop-filesystem-client-aws, hadoop-filesystem-client-openstack, etc, each with their own tests, JARs and POM file dependencies. This would translate into separate bigtop RPMs/JARs