Research On Transaction Management For Massive Data In NOSQL

With the development of the information technology, the explosivegrowth of data leads to the problem of massive data storage and processing.The traditional relational databases are not very effective when dealing withthe high concurrent reading and writing of massive data and not to mentionthe well-known weak support of high scalability. NoSQL meets aboverequiements, which makes it very popular in cloud environment, wherescalability requirement is very high. However, in contrast to traditionalrelational database, NoSQL makes a compromise in data consistency.In order to solve the problem of low data consistency of NoSQL, thispaper proposed the transaction support mechanism for massive data inNoSQL illustradted by the case of HBase. Relying on the self-providingfeatures of HBase, such as persisting, high throughput, low latency, the lockservice is implemented by a single row in HBase. Two phase commitprotocol is used to synchronize multi-rows in the transaction. Exceptionhandling mechanism is also provided to handle the problem of nodecrashing, which is quite common in distributed systems. Three differentisolation levels (read uncommitted, read committed and repeatable read) arerealized to satisfy different requirements of transaction isolation in varioussystems.On this basis, a futures trading simulation system is established toverify our transaction support mechanism. The futures trading simulationsystem uses HBase as its database, and is capable to cope with thousands ofconcurrent queries from hundreds of simulated traders. The outcome ofverification shows that after applying the transaction support mechanism, NoSQL can meet the requirement of strong consistency from large scaleconcurrent access. Although transaction support mechanism leads to adecline in efficiency, but the decline is still in an acceptable range.