The Design And Implementation Of Multi-row Transaction In HBase System

With the rapid development of information distributed in cloud computing,Hadoop is becoming more and more perfected. It has been handled more in Internet.HBase which belongs Hadoop has has excellent preformance and strong expansibility,but, some problem there, it does not support multi-row transaction functions and stillworry many users. Current solution is that users develop there own Multi-rowTransaction of application in HBase. In that way, the solution can not open. Not onlyincreasing the cost of the software development cycle, but also impeding HBase beingpopular in the inter industry and optimization. Along with the upgrade version of theHBase itself, many practical interface functions have been added to the project code,such as interface Coprocessor. Although the current system supports only the controlof single data transaction control, but if we can make good use of these highlyextensible interface, Combine with the existing HBase system function and reliaze themulti-row transaction function that user expect, which is the main contents of thispaper.According to the character of HBase system architecture and the transaction oftraditional database, multi-row transaction in HBase has been designed and realized.The system maintains the HBase itself extensibility, high throughput, availabilitycharacteristics. Useing Hazelcast which is a distributed memory system as cache areacan ensure the reliability of data and the I/O throughput of transaction logs.Analysising the transaction processing algorithm and mechanism of relationaldatabase with the structure of HBase system can add the appropriate concurrencycontrol and resource lock algorithm into it. In as far as possible in the smallest scopemodify the original processing logic. Using the interface Coprecessor and Filterdesigns the function of beginning, commit and rollback. Through the implementationof Index function reflect that the transaction is important in HBase.Through the performance test, it compare the multi-row transaction HBase andthe original system in tme consuming and resource loss. Confirming the indextransactional function correctly and getting the conclusion prived that the system hasfavorable function and performance is meeting the requirements of multi-rowtransaction in HBase.