Locking protocols between transactions have long been an important field in database research. In a classic main memory database system, transactions tend to be short and small so that the system usually adopts some simple locking protocols like single-write-multiple-read. Transactions are executed in an almost serialized sequence. In a system of heavy transaction concurrencies but relatively small overlapping data sets between transactions, locking protocol could be improved by introducing coarse-grained intent share lock and lock granularity transformation in order to boost concurrencies between transactions, raise transaction throughput and improve overall system performance.