Optimization Processing Techniques For Multi-tenant Query Workload In Data Market

As the value of data has received more and more attention,the data market has gradually been accepted by more users.The data market uses a new cloud-service model called Daa S(Data-as-a-Service)and users do not need to consider the hardware and software resource configurations.All they have to do is to find and subscribe the data set they interested in the market platform,and then they can acquire the corresponding results.Service providers need to provide appropriate hardware resources and software services to maintain and guarantee tenants' service requests.In order to reduce costs,they need to increase the resource reuse rate,so that limited system sources can serve more users at the same time.However the reuse of resources may reduce the quality of some tenant services.To ensure the quality of service in the data market,each tenant will reach the SLA(Service Level-Agreement)with the data service provider,specifying the payment method and the corresponding service price model.In order to obtain revenue through its own service,it must ensure that the quality of the service ultimately obtained by the tenant,otherwise providers will be subjected to corresponding penalties for breaching the agreement.How to effectively optimize services,improve system resource utilization and tenant query processing efficiency to maximize platform revenue has become a key issue for service providers.The service provider wants to meet the service requirements of more tenants with less cost.When the total resources are shared by multiple tenants,one tenant occupies part of the resources,which will cause other users' services and overall revenues to be affected.This paper takes the service provider's revenue maximization as the criterion and considers the characteristics of tenant's demand.And it can study and discuss the multi-tenant load integration and query scheduling.The main contributions for this paper are as follows:1.A price model that better meets the needs of tenants and service providers is proposed.In the new price model,charges are based on the ”quiry time required for the amount of data per unit”.To maximize the revenue,the service provider needs to consider the cost time of the tenant query and the size of the subscripted data set.The model not only takes into account the speed of query execution,but also considers the occupancy of the system resources by the data set.For tenants,the new price model allows them to better understand the quality of their services and know the relative processing speed of submitted their queries.2.A query load integration mechanism based on dynamic programming is proposed.In the Daa S,all the workloads 'grab' whole system resource,so increasing the system resource reuse rate can be achieved by efficiently integrating the tenant workload.This paper combines the price model and considers the different performance of different database engines for the same workload based on the tenant behavior and historical analysis of the load.Through the dynamic programming method,system could find the most valuable load set for system revenue and integrate this part of the load into the memory-database,and integrate the remaining load into the disk-database.3.A query scheduling mechanism based on sliding window is proposed.During the service process,when the query queues for system execution,all of the queries needed to be reasonably scheduled.On the one hand,it can improve the utilization of resources and allow limited resources to serve more tenants.On the other hand,it can reduce the wait time of queries and maximize the benefits of the service provider.Based on the sliding window,this paper fully considers the queuing request falling within the sliding window and the possible expected revenues,then generates a scheduling sequence to ensure the system revenue and executes the tenant query as quickly as possible.4.A Daa S workloads processing system prototype is constructed based on the proposed price model and algorithms.At the same time,using the TPC-H and TPC-DS benchmark data sets,benchmarked the effectiveness of the proposed methods.In conclusion,this paper studies the multi-tenant service problem in the data market and studies the multi-tenant workloads integration and queries scheduling technology.Through the workloads integration mechanism of dynamic programming and the query scheduling mechanism based on sliding window,the performance of multi-tenant service is improved,and the benefits of service providers are enhanced.And the corresponding data sets are provided for testing,which proves that the methods proposed in this paper can achieve good results in the application.