A Cloud Controller for Performance-Based Pricing

Abstract

Abstract-New dynamic cloud pricing options are emerging with cloud providers offering resources as a wide range of CPU frequencies and matching prices that can be switched at runtime. On the other hand, cloud providers are facing the problem of growing operational energy costs. This raises a trade-off problem between energy savings and revenue loss when performing actions such as CPU frequency scaling. Although existing cloud controllers for managing cloud resources deploy frequency scaling, they only consider fixed virtual machine (VM) pricing. In this paper we propose a performance-based pricing model adapted for VMs with different CPU-boundedness properties. We present a cloud controller that scales CPU frequencies to achieve energy cost savings that exceed service revenue losses. We evaluate the approach in a simulation based on real VM workload, electricity price and temperature traces, estimating energy cost savings up to 32% in certain scenarios. I. INTRODUCTIONWith the wide range of VM types, heterogeneous infrastructure and different computing environments including VMs and containers [1], estimating the performance of provisioned resources is becoming increasingly challenging [2]. New cloud pricing schemes are emerging where resources are priced based on the delivered performance. For example, the CPU frequency provided to a VM at runtime determines the price [3], with high CPU frequencies being more expensive. We call this model performance-based pricing and it is used in production by cloud providers, such as ElasticHosts [4]. Though this approach mainly targets users, it could also be used by cloud providers to control their energy consumption. Energy consumption of data centers is becoming a major issue, accounting for 1.5% of global electricity usage [5]. Furthermore, modern clouds may consist of geographically-distributed data centers influenced by dynamic local factors, such as real-time electricity prices [6] and temperature-dependent cooling [7], that we call geotemporal inputs.We call the subsystem of the cloud, that determines the actions to allocate and manage the VMs, a cloud controller. Adapting the cloud controller to geotemporal inputs through actions, such as CPU frequency scaling, raises a trade-off problem between the potential energy savings and service revenue losses incurred under performance-based pricing. This is the challenge at the core of this paper.Frequency scaling is a power management technique commonly used to lower the operating frequency of hardware resources in order to reduce power consumption [8]. However, frequency reduction may degrade the performance of resources. Depending on the workload characteristics, workload

Managing energy, performance and cost in large scale heterogeneous datacenters using migrations

“…Furthermore, an experimental study is conducted to investigate the effect of placement decisions, when several VMs are placed on same core/host or neighbouring hosts. In [25], the authors investigated several scheduling policies combined with a consolidation technique to reduce the energy cost which is based on VMs performance level. The authors proposed a performance-based pricing model to increase service revenue and decrease the system energy consumption that can be up to 32%.…”

“…It is necessary to quantitatively determine the balance between energy savings and workload performance to find the optimal number of VM migrations. Current work has explored the hardware heterogeneity and performance variation of different instance classes, however, the effect of scheduling techniques and VM migrations is not addressed, when different applications are taken into account -and with the notable exception of [24], [25], [26], this is rarely addressed.…”

Allocating Limited Resources to Protect a Massive Number of Targets Using a Game Theoretic Model

“…Meanwhile, since user requests are becoming necessary while providing service, some research has focused on user-demanddriven resource allocation [26], [27], [28]. Later, the costoptimized and performance-based schemes are combined to allow a resource provider to achieve a win-win objective in which resource provider obtains the maximum profit while the user receives the best experience [29], [13], [30], [31].…”

Dynamic energy-aware scheduling for parallel task-based application in cloud computing

“…The second strategy, a complement to the first strategy, reduces the monetary costs of non-critical tasks. This algorithm shows that we can reduce monetary costs while producing a good makespan like [17] does, but without using the DVFS technique.…”

“…In [17] the authors propose a cloud controller for performancebase pricing; the cloud controller is in charge of determining actions to allocate and manage the VMs, taking into account CPU frequency scaling, pricing and geo-temporal inputs such as real-time electricity prices [18] and temperature-dependent cooling [19], the problem is balancing the trade-off between energy savings and revenue loss for performing actions by scaling CPU frequency, for creating and migrate, suspend or resume VMs that have to pay a high cost in energy. The proposed controller can determine the optimal CPU frequency between energy savings and workload performance, and can reevaluate control actions at run-time when considering geo-temporal inputs, electricity prices, and temperatures.…”