Research Of Concurrency Control In DDBS Based On Lazy Replication

Lazy replication is a popular and useful technique for improving the performance and availability of distributed database systems. Owing to it's accessing local data and avoiding to access remote data using redundancy, it can not only reduce communication cost and response time, but also run system rightly even in some sites crashed and communication line failure. In lazy replication database systems, the key issues are: how to keep the databases consistency and synchronization for Primary and Secondary sites. Although there are concurrency control techniques which guarantee serializability in lazy replication systems, these techniques may result in undesirable transaction orderings. Transactions may see stale data, because they may be serialized in an order different from the one in which they were submitted. Strong serializability avoids such problems, but it is very costly to implement and even reduce systems'performance.In this thesis, we use concurrency control techniques based on timestamp, to implement a generalized form of strong serializability (Strong Session 1SR) which is suitable for lazy replication. This is accomplished by grouping transactions into sessions. Transaction ordering is preserved between transactions in the same session and avoids"stale data"between these transactions, but not between transactions in different sessions. The primary site generates the only timestamp for the transactions and prorogating refresh transactions lazily to secondary sites in serialization order. Reducing communications volume between sites by making full use of the control information stored in database items. With these, in addition to having many of the advantages of strong serializability, it not only can avoiding"stale data", and also can improve the performance and concurrency rate of the total system.There have been some proposals for scaling-up a database system using lazy replication. In these proposals, system scale-up is achieved through the addition of secondary sites. The addition of more secondary sites improves system performance of read-only transactions. However, the scalability of the distributed system is limited by the single primary site which services the update workload. In this thesis, we present an approach to scale-up the primary sites: Each primary site only has some part of main data of the system. We use L-Vectors (Lexicographically Ordered Vectors) to record the number of transactions committed in each primary site, then the L-Vectors can stand of the state of each primary site, guarantee data consistence under lazy approach between each primary site. And guarantee Strong Session 1SR under concurrency control techniques based on timestamp for the whole system. This model and techniques provide good scalability for primary site and secondary sits, and also avoid"dirty data", which can occur when transactions access stale replications.