No abstract

Read separating from write is a strategy that NewSQL adopts to incorporate the advantages of traditional relation database and NoSQL database. Under this architecture, baseline data is split into multiple partitions stored at distributed physical nodes, while delta data is stored at single transaction node. For reducing the pressure of transaction node and improving the query performance, delta data needs to be synchronized into storage nodes. The current strategies trigger the procedure of data synchronization per partition, meaning that unchanged partitions will also participate in data synchronization, which consumes extra network cost, local IO and space resources. For improving the efficiency of data synchronization meanwhile mitigating space utilization, the fine-grained data synchronization strategy is proposed, whose main idea includes that fine-grained logical partitions upon original coarse-grained partitions is established, providing more correct synchronized unit; the delta data sensing strategy is introduced, which records the mapping between changed partitions and its delta data; instead of partition driven, the data synchronization through the delta-broadcasting mechanism is driven, constraining that only changed partitions can participate in data synchronization. The fine-grained data synchronization strategy on Oceanbase is implemented, which is a distributed database with read separating from write, and the results show that our strategy is better than other strategies in efficiency of data synchronizing and space utilization.

No abstract

The traditional lock manager (LM) seriously limits the transaction throughput of the main memory database systems (MMDB). In this paper, we introduce dependence-cognizant locking (DCLP), an efficient improvement to the traditional LM, which dramatically reduces the locking space while offering efficiency. With DCLP, one transaction and its direct successors are collocated in its context. Whenever a transaction is committed, it wakes up its direct successors immediately avoiding the expensive operations, such as lock detection and latch contention. We also propose virtual transaction which has better time and space complexity by compressing continuous read-only transactions/operations. We implement DCLP in Calvin and carry out experiments in both multicore and shared-nothing distributed databases. Experiments demonstrate that, in contrast with existing algorithms, DCLP can achieve better performance in many workloads, especially high-contention workloads.

Data processing can be roughly divided into two categories, online transaction processing OLTP(on-line transaction processing) and online analytical processing OLAP(on-line analytical processing). OLTP is the main application of traditional relational databases, and it is some basic daily transaction processing, such as bank pipeline transactions and so on. OLAP is the main application of the data warehouse system, it supports some more complex data analysis operations, focuses on decision support, and provides popular and intuitive analysis results. As the amount of data processed by enterprises continues to increase, distributed databases have gradually replaced stand-alone databases and become the mainstream of applications. However, the current business supported by distributed databases is mainly based on OLTP applications, lacking OLAP implementation. This paper proposes an implementation method of HTAP for distributed database CBase, which provides an implementation method of OLAP analysis for CBase, and can easily deal with data analysis of large amounts of data.