A Study On Resource Allocation Under Saas Model With Multi-tenant’s Uncertain Demand

SaaS (Software as a Service) reduces the cost of both information serviceproviders and users, and can provide personalized services to the mass SaaS tenants.Through purchasing SaaS services, SMEs can effectively improve the degree ofinformation, thus SaaS is an important approach for promoting development ofenterprises’ information. With the industrial development of SaaS, how to reduceservice cost and energy consumption has become an essential problem. However, thetraditional approaches for SaaS resource allocation are based on certain users’ demand.There are few methods focusing on effectively allocating resource based on massiveuncertain demand under cloud environment. The traditional approaches are likely tocause error of demand between users and providers. Error of individual tenants issmall, but the total cost of all error is enormous. Therefore, how to accurately predicttenants’ real demands and develop corresponding optimal allocating schemes is thekey research problem in this paper.In order to predict tenants’ uncertain demand, a SaaS tenants’ model, based onSLA, is built, which contains description of tenants’ demand, matching rules andrequesting approaches. In existing SaaS mode, SLA, signed by tenants, cannot bechanged during the contract period. But a tenant consists of many internal users,which determines the demands of tenants are dynamic. In existing SaaS mode, basedon fixed SLA, a large part of the service cost is not used by tenants. Therefore, inorder to simulate tenants’ really uncertain demand, the BP neural learning networkprocess is used to simulate SaaS service data. And a new service request mode, inwhich SLA can be changed, is proposed to achieve the goal of “pay as you go”indeed.To achieve the ultimate goal, working out schemes of resource allocation in time,which can adapt users’ uncertain demand, a two-goal SaaS resource allocating modelis built. In this model, tenants’ demands are converted to the form of resource.Through searching for the best allocating plan for demands of resource, the task ofallocating tenants’ service request by using virtual machines is completed. Considering the difficulty of solving this problem, a new resource optimal allocatingalgorithm based on swarm intelligence is proposed. This algorithm focuses onsearching optimal resource allocating scheme, which takes cost and service defaultrate as fitness. Compared to other algorithms, this algorithm is more adapted to solvethe problem due to its high performance and low complexity. In addition, taking thecharacteristics of massive tenants and dynamical demands into account, the algorithmis improved through decomposing tasks and mutating variations. As a result, theresource allocating algorithm can run in parallel by using the free resources ofreleased virtual machine. Furthermore, both the quality of schemes and the speed ofalgorithm are enhanced. Finally, simulating experiments aimed to test theeffectiveness of algorithms and models are implemented. The results show thatoptimal resource allocating schemes can be searched out in time by using the modeland the corresponding algorithms proposed in this paper.Recognizing uncertain demands the approach makes the goal of predictingtenants’ real demands possible. This approach reduces the errors of SaaS servicematching processes. Correspondingly, a service request model in which SLA can bechanged is proposed to make SaaS more practicable. Moreover, an optimal SaaSresource allocating model facing tenants’ uncertain demands is built to guarantee thequality of service and reduce the total cost. In addition, this paper explores newalgorithms for searching schemes of allocating SaaS resource in cloud environment.This study provides foundation of allocating SaaS resource efficiently.