Finer grained concurrency for the database cache

The database cache transaction recovery technique as proposed in [Elhard and Bayer 84] offers significant performance advantages for reliable database systems. However, the smallest granularity of locks it provides is the page. Here we present two schemes supporting smaller granularity. The first scheme allows maximal concurrency consistent with physical two-phase locking, with the same per-transaction I/O cost as original database cache scheme. The second scheme offers the same concurrency as the first, but features reduced I/O on commit, at the cost of some increase recovery time.