Database Evaluation Under High Contention Scenarios

In recent years,new kinds of applications have sprung up.In order to provide new services or ensure the quality of services,these applications usually put forward new demands on the functions of database systems,such as large-scale data storage,high concurrent data access,high resource competition,etc.Therefore,traditional database systems constantly face new business and hardware,and launch the architecture adjustment or design optimization.Additionally,the distributed platform breaks the closure deployment environment of the traditional database system,such as cloud cluster.They have the characteristics of expandability,high flexibility,and easy deployment,which provide a good environment for the database system,but also bring new problems for stable service quality of the database.With the new requirements of applications,new technologies of database systems,and increased complexity of deployment environments(cloud platforms),we are challenged by designing and implementing an effective benchmark for database systems to support quality control along with database development for new applications.Database benchmark has always been an important guarantee for the healthy development of the database industry,which ensures fair competition among different database manufacturers.Benchmarks typically abstract a class of applications,build representative data and workloads,and define a set of performance metrics.Compared with the traditional benchmark workload,first of all,Seckill represents a typically new type of database application.The workload of this kind of application is characterized by high dynamic,high concurrency,and high contention.However,traditional benchmarks cannot simulate these new application scenarios.Secondly,the lack of quantitative transaction contention control means lead to inadequate evaluation of the transaction processing capability of database systems.Moreover,resource changes in the open environment,e.g.CPU fluctuations,affect the evaluation results.Dynamic environment simulation is a necessary means to ensure the validity and reference of database evaluation.Given the above problems,this thesis defines the benchmark for new applications of Seckill.Besides,this thesis implements an extensible quantitative method to control workload contention.Moreover,this thesis constructs the correlation model among the environment features and designs the algorithm of dynamic simulation of the runtime environment.The main work and contributions are summarized as follows:(1)Propose a benchmark for Seckill applications(Peak Bench).Targeting the Seckill scenario,this thesis abstracts the business model and proposes a new benchmark.The workload has the characteristics of high concurrency,high contention,and high dynamic.This is the first work to provide a transaction contention representation and realize fine-grained contention control.This work defines the dynamic and multidimensional skew characteristics of transaction workloads to compensate for the inadequacy of traditional benchmarks.For the usage of Peak Bench,this thesis develops a supporting tool,which can be easily configured and adapted for simulating Sec Kill scenario.The usability of the benchmark is verified by a large number of experiments compared with the existing representative benchmarks.(2)Design scalable contention control ways for transactional workloads.This thesis designs and implements an extensible workload contention control method,which simulates transaction contention quantitatively based on two iterative algorithms with different contention granularity.The traditional method uses the thread waiting model to simulate the contention.It cannot simulate large-scale real-time transaction contentions and has a scalability problem.According to the performance state(i.e.,Latency)of databases,this work adjusts the access distribution of data in real-time,generates contention workload,and implements extensible contention control.Through abundant experiments,the feasibility and effectiveness of the proposed method are verified.Under the control of fine-grained contention,this work summarizes the processing capabilities of the typical open-source databases for supporting support of the high contented transactions.(3)Design a simulation algorithm for dynamic application environment.This thesis presents and designs a method for environment simulation of database systems.First,this work defines four types of parameter-sensitive workloads facing key environmental features.Then,a multiple-order polynomial regression model is established according to the consumption of environmental resources and the interaction of workload on a single dimension.Furthermore,this work establishes a multivariate linear regression model for the complex correlations among different dimensions of the environment.Finally,based on the environment model,this work designs a dynamic environment simulation algorithm.The effectiveness of this work is verified by the environment simulation experiment and the results demonstrated on the opensource database systems.It launches experiments to demonstrate the influence of the environment on the performance of Sec Kill applications.In summary,to deal with the difficult problems in performance evaluation of transactional database systems,this thesis proposes a new benchmark for the Seckill applications,realizes the fine-grained contention control of transaction workloads,and achieves modeling running environment.To support the use of Peak Bench,it develops a set of tools.The correctness and effectiveness of the proposed method are proved by a large number of experiments on typical open-source database systems.The research results of this thesis can reduce the implementation cost of database evaluation,provide means for the selection of database systems for seckill applications,and promote the technical development and fair comparison of the database system.In future work,we will further optimize the supporting tools,improve the execution efficiency of the tools.We will launch the assessment to typical open-source database systems and provide detailed evaluation reports.Finally,we will make the tools open source to promote the use of Peak Bench.