Study On Software Transaction Memory Under Multi-thread Condition

Multiple processing cores being integrated into a chip become the most important way of improving processor's performance. It can decrease the power cost and increase multi-thread computing performance of processor, without increasing the processor's frequency. However, to make full use of multi-thread computing performance in multi-core processors, the most difficult challenge is how to build the multi-thread computing model. Today, most multi-thread programs use lock to promise their synchronization. But the current lock-based abstractions are difficult to use and make it hard to design computer systems that are reliable and scalable, and the program are hard to understand with low reusable. To address some of these difficulties, STM( software transaction memory ) is invented to perform groups of memory operations atomically. Using transactional memory instead of locks brings well-known advantages: freedom from deadlock and priority inversion, automatic roll-back on exceptions or timeouts , and freedome from the tension between lock granularity and concurrency.At now, the study of STM focuses on how to speed up the commit time for each transaction and the priority algorithm for choosing which transaction to perform. But few work on coordinating and communicating between transactions. On this situation, this paper introduces a STM model called CSTM which through gives the transactions ability to block themselves and composing them into a new transaction( both sequence and select ), offers coordination and communication method between transactions and make transactions functions more reusable. The result of the test shows in the condition where transactions interactions, CSTM gains higher performance against lock-based strategy and other STM systems with simple transaction methods.