Research And Implementation Of Serializable Isolation Level For Composite Workloads

With the development of the Internet,big data,and artificial intelligence,modern enterprise-level applications not only need to support traditional high-concurrency OLTP services,but also often require efficient real-time data analysis.This mixed transaction load is typically called HTAP(Hybrid Transactions and Analytics Processing)workload.The HTAP workload contains read/write requests on a single record with primary key and key-range scan queries.To support high concurrent transaction access,many modern applications use in-memory databases running on multi-core,large-memory servers to achieve high throughput and low latency.Although most large data systems reduce concurrent concurrency constraints in order to improve concurrency,serialized isolation levels are still required in certain financial systems to ensure the correctness of the business.Most of the current in-memory databases do not support HTAP requests well.In order to ensure serializable execution of transactions,concurrency control strategies spend a lot of time on the key-range scan request for conflict detection.Therefore,we designed and implemented an efficient and scalable optimistic concurrency control strategy(ROCC)based on logical scope verification.By dividing the logical interval,the transaction verification at the range level reduces the cost of transaction verification and improves the overall performance.We also use an optimized data structure — lock-free circular arrays — to track changes of transactions within each logical partition,perform read and write collision detection,and reuse transaction memory.We have proved through analysis and experiment that ROCC can effectively reduce the verification cost.Compared with other methods under HTAP workload,ROCC has better performance and scalability.In addition,we also applied ROCC to an open source distributed system,CEDAR.We described specific implementation methods on CEDAR for logical range verification to efficiently support executing transactions under the serializable isolation level in a HTAP workload.In this paper,we specifically made the following contributions:1.We proposed optimistic concurrency control algorithm(ROCC)based on Logic Range Verification(RV).The algorithm can be used to verify the scan load in the logical scope granularity,avoid the verification of the scanned data one by one,and can effectively support mixed load of OLTP and range queries.2.To support efficient range validation,we designed an efficient transaction management strategy within logical partitions.3.We conducted extensive experiments on a prototype system of an in-memory database and proved the efficiency of ROCC under various loads.4.We applied RV to a real distributed database system and conducted experimental tests.