Hierarchical database decomposition: a technique for database concurrency control

• Authors: Meichun Hsu;Stuart E. Madnick
• Venue: PODS '83 Proceedings of the 2nd ACM SIGACT-SIGMOD symposium on Principles of database systems
• Year: 1983

The classical approaches to enforcing serializability are the two-phase locking technique and the timestamp ondering technique. Either approach requires that a read operation from a transaction be negistered (in the form of either a read timestamp or a read lock), so that a write operation from a concurrent transaction will not interfere improperly with the read operation. However, setting a lock or leaving a timestamp with a data element is an expensive operation. The purpose of the current research is to seek ways to reduce the overhead of synchronizing certain types of read accesses while achieving the goal of serializability.To this end, a new technique of concurrency control for database management systems has been proposed. The technique makes use of a hierarchical database decomposition, a procedure which decomposes the entire database into data segments based on the access pattern of the update transactions to be run in the system. A corresponding classification of the update transactions is derived where each transaction class is 'rooted' in one of the data segments. The technique requires a timestamp ordering protocol be observed for acesses within an update transaction's own root segment, but enables read accesses to other data segments to proceed without ever having to wait or to leave any trace of these accesses, thereby reducing the overhead of concurrency control. An algorithm for handling ad-hoc read-only transactions in this environment is also devised, which does not require read-only transactions to wait or set any read timestamp.