Cross-cloud Collaborative Optimization For Interactive Services

In recent years,the cloud computing has grown rapidly.It also brings two major problems: Data centers belonging to multiple CSP(Cloud Service Providers)provide different resource price,which causes the operation cost of the cloud service varies wildly;Data center load become huge,and the average cooling energy consumption accounts for about 50% of the total energy consumption,resulting in the energy saving problem becomes more significant.The two problems can be solved with cross-cloud collaborative optimization algorithm,but the algorithm must ensure the cloud service met the tail latency constraint after rescheduling.This thesis tries to overcome the challenges with two scheduling algorithms.The first algorithm solves a two-stage SLP(Stochastic Linear Programming)problem to reduce the operation cost.In stage one,it chooses the lowest price data center group.In stage two,it uses decomposition technique to adjust the resource provision plan to handle the demand uncertainty.The second algorithm solves a proactive weather-aware optimization function to reduce cooling energy consumption.The tail latency constraint is estimated with the G/G/M queue model.In summarize,the contributions of this thesis include:1)A cost model comprises the reservation resource and the on-demand resource for multi-CSPs is proposed.Importantly,complex distribution decision and resource provision planning are achieved by modeling tail latency constraint with SLA(Service Level Agreement).2)With resource price uncertainty,the algorithm chooses the lowest resource price data center with SLP coarse-grained.With resource demand uncertainty,the algorithm estimates the resource demand through historical execution;and uses decomposition technique to develop fixed resource provision plan in target data center group fine-grained.3)The cooling energy consumption model considers the energy for reducing the heat conducted with outside world and the heat generated by the IT equipment.CDD(Cooling Degree Day)is used to formulate the cooling energy consumption.4)An algorithm is proposed to make scheduling decisions for the load balance.It dynamically schedules the workloads to data center with lowest temperature.This work is the first that takes a holistic approach by covering the cooling energy consumption of the data centers from CDD to adjust the workloads scheduling in time slot.