In Cassandra 0.7, keyspaces and column families may be created or dropped in a live cluster. In 0.6, the procedure is a bit more difficult.
Before you can remove or add column families, you must clear the commit log. To do this, run nodetool drain on the node; this will cause the node to stop accepting writes, flush the memtables to disk, and clear the commit log. When this has completed (which you may tell by checking that all flush stages have no active or pending operations as shown by nodetool tpstats), you may stop Cassandra.
After draining and stopping Cassandra, simply add the column family definition to storage-conf.xml and start Cassandra. You should wait until all nodes have done this before using the column family with clients.
After your clients are no longer using a column family, you may remove it from each node in the cluster one-by-one.
After Cassandra has been drained and stopped, delete all of the data files associated with the column family. These include files of the form CF-VERSION-Data.db, CF-VERSION-Index.db, and CF-VERSION-Filter.db, where CF is the name of the column family you are deleting.
Next, remove the column family definition from storage-conf.xml and restart Cassandra.
It is not recommended that you attempt to rename a column family.