1) What are the things that one should practice to maintain Cassandra cluster properly?
2) As explained here:Apache Cassandra at Clearspring, they are doing rolling restart in a week or two.
3) How does rolling restart help in maintaining Cluster? Is there any specific activities (Cleaning) that Cassandra does while restarting?
The only real generic rule is that you must run nodetool repair at least every gc_grace_seconds to avoid deleted data reappearing. Otherwise, general monitoring of your cluster using cfstats, iostat, etc., is always a good idea.