基于LSA-STM的读集校验优化方法

AN OPTIMISATION METHOD OF READ SET VALIDATION BASED ON LSA-STM

在基于对象、基于时间、无阻塞的LSA-STM中,读集的校验开销随着事务读取数据对象的数量而成线程增长。为降低事务读集的校验开销,提出读集分片校验策略(RSV)。将读集及数据对象分片,并引入分片序列来动态记录数据片的变化。校验读集时通过检测数据片的变化来决定是否跳过相应读集分片的校验。经实验表明,在读集庞大、冲突事务较少的情况下,该策略可提升事务吞吐量90%左右。

In object-based,time-based and non-blocking LSA-STM,the validation cost of read set increases in the form of thread along with the numbers of data objects the transactions read. For reducing the validation cost of the transaction read set,we present the read set slicing validation（ RSV） strategy. RSV algorithm divides the read set and data objects into slices and introduces the slicing sequence to record the variation of the data slice dynamically. When validating the read set,transaction can decide whether to skip or not the validation on corresponding read set slice by checking the change of the data object slice. It is demonstrated by the experiment that in the case of huge read set and few conflict transactions,this strategy can increase the throughput of transactions by about 90%.