KRISHNA - An efficient concurrency control algorithm based on dynamic attributes of transactions and its performance

Concurrent transactions during their execution life acquire a number of attributes. We call them dynamic attributes of transactions. We have used them to develop a concurrency control mechanism, called KRISHNA, using a two-phase locking policy. To establish the usefulness of dynamic attributes in developing concurrency control mechanisms and to investigate the behaviour of KRISHNA we have studied their performance using a detailed simulation model. We show that KRISHNA outperforms a number of well-known concurrency control mechanisms based on a two-phase policy. We also show that the conflict-resolution policy of KRISHNA is comparatively more accurate in resolving conflicts, i.e., it selects the 'right' transaction for rolling back or blocking. We argue that implementing KRISHNA is not expensive compared to commonly used concurrency control mechanisms.