Dynamic computing resource adjustment for enhancing energy efficiency of cloud service data centers

Abstract

Cloud computing clusters distributed computers to provide applications as services and on-demand resources over Internet. From the perspective of average and total energy consumption, such consolidated resource enhances the energy efficiency on both clients and servers. However, cloud computing has a different power consumption pattern from the traditional storage oriented Internet services. The computation oriented implementation of cloud service broadens the gap between the peak power demand and base power demand of a data center. A higher peak demand implies the need of feeder capacity expansion, which requires a considerable investment. This study proposes a computation related approach to lessen the increasing power demand of cloud service data centers. Through appropriated designs, some frequently used computing algorithms can be performed by either clients or servers. As a model presented in this paper, such client-server balanced computation resource integration suggests an energy-efficient and cost-effective cloud service data center.

A Survey on Modeling Energy Consumption of Cloud Applications: Deconstruction, State of the Art, and Trade-Off Debates

“…The bigger speedup might indicate the better offloading opportunity, and lead to the higher application performance and the lower energy consumption [42,41,70]. 3) CPU Utilization: The studies [53,29] considered the power consumption in a server to be an exponential function of its CPU utilization, and the high CPU utilization is related to the underlying large workload size. Accordingly, higher utilization would result in more energy consumption within the same size of time window [88].…”

“…5) Task Complexity: The computational complexity in tasks or functional modules is closely associated with the Cloud application's energy consumption [68,97,29], as complex computation requires more computing resources and/or causes longer execution time. To verify this association, the empirical studies varied task complexity mainly through topping up functions [53] and increasing the load of mathematical calculations [81,56], while the simulation study [37] characterized the complexity in computation algorithm as a random variable with Gamma distribution. 6) Task Size: As mentioned previously, a composite-object task can further be defined as a combination of the input/output data and computation workload [41], and therefore the size of a task can partially be reflected by the data size [57] or together with the computation complexity [37].…”

“…Therefore, we particularly label Equation (19) as an active energy consumption model. Instead of a constant value, the power consumed in a compute resource has been identified to be an exponential function of the resource utilization [53]. By using P idle (r cpu ) and P full (r cpu ) to respectively represent the compute resource's empty and full load powers, the energy consumption for running the task n cpu on the compute resource can be modeled as: 20) where α and β are resource-specific parameters that need to be determined through empirical measurements.…”

“…As for a multi-CPU server, U (t) was estimated as the number of active CPU cores among all the available ones [91]. Considering that the compute resource utilization would also be proportional to the workload being dealt with, the study [53] further modeled U (t) = γ · W (n cpu , t) + λ, where γ and λ are both resource-specific parameters, and the workload W (n cpu , t) was measured by the number of user connections at time t.…”

“…Moreover, to intentionally study the mutual actions between a Cloud application and its users, it would be useful to distinguish interaction from the other types of communication elements. For example, instead of reflecting communication data throughput, this execution element is often highlighted when stressing the server load, like user connections [53] and user requests for playing online games [33] or for exploring HTTP websites [54,55].…”