Chase et al. [52] have studied the problem of managing resources in Internet hosting centers. Resources are shared among multiple service applications with specifiedSLA—the throughput and latency. The authors have developed an OS for an Internet hosting center (Muse) that is a supplement for theOSsof the individual servers and supposed to manage and coordinate interactions between the data center’s components. The main distinction from previous research on resource management in hosting centers is that the objective is not just to schedule resources efficiently but also to minimize the consumption of electrical power by the system components. In this study, this approach is applied to data centers in order to reduce: operating costs (power consumption by computing resources and cooling system); CO emissions, and thus the impact on the environment; thermal vulnerability of the system due to cooling failures or high service load; and over-provisioning in capacity planning. Muse addresses these problems by automatically scaling back the power demand (and therefore waste heat)
when appropriate. Such a control over the resource usage optimizes the trade-off between the service quality and price, allowing the support of flexibleSLAnegotiated between consumers and the resource provider.
The main challenge is to determine resource demand of each application at its current request load level, and to allocate resources in the most efficient way. To deal with this problem, the authors apply an economic framework: the system allocates resources in a way that maximizes the “profit” by balancing the cost of each resource unit against the estimated utility, or the “revenue” that is gained from allocating that resource unit to a service. Services “bid” for the resources in terms of the volume and quality. This enables negotiation of the SLA according to the available budget and current QoS requirements, that is, balancing cost of resource usage (energy cost) and benefit gained due to the usage of this resource. This enables a data center to improve the energy efficiency under a fluctuating workload, dynamically match the load and power consumption, and respond gracefully to resource shortages.
The system maintains an active set of servers selected to serve requests for each service. Network switches are dynamically reconfigured to change the active set when necessary. Energy consumption is reduced by switching idle servers to power- saving states (e.g., sleep, hibernation). The system is targeted at the web workload, which leads to a “noise” in the load data. The authors address this problem by applying the statistical “flip-flop” filter, which reduces the number of unproductive realloca- tions and leads to a more stable and efficient control.
This work has created a foundation for numerous studies in the area of power-
efficient resource allocation at the data center level; however, the proposed approach has several weaknesses. The system deals only with the CPU management, but does not take into account other system resources such as memory, disk storage, and network interface. It utilizes APM, which is an outdated standard for Intel-based systems, while currently adopted by industry standard is ACPI. The thermal factor is not considered as well as the latency due to switching physical nodes on/off. The authors have pointed out that the management algorithm is stable, but it turns out to be relatively expensive during significant changes in the workload. Moreover, heterogeneity of the software configuration requirements is not handled, which can be addressed by applying the virtualization technology.
8.2.3 Energy-Efficient Server Clusters
Elnozahy et al. [20] have explored the problem of power-efficient resource management in a single-service environment for web applications with fixedSLA(response time) and automatic load-balancing running in a homogeneous cluster. The motivation for the work is the reduction of operating costs and improvement of
the error-proneness due to overheating. Two power management mechanisms that
are applied switching physical nodes on and off (vary-on vary-off, VOVO) and
DVFS of the CPU.
The authors have proposed five policies for the resource management: indepen- dent voltage scaling (IVS), coordinated voltage scaling (CVS), VOVO, combined policy (VOVO-IVS), and coordinated combined policy (VOVO-CVS). The last mentioned policy is stated to be the most advanced and is provided with a detailed description and mathematical model for determining CPU frequency thresholds. The thresholds define when it is appropriate to turn on an additional physical node or turn off an idle node. The main idea of the policy is to estimate total CPU frequency required to provide the expected response time, determine the optimal number of physical nodes, and proportionally set the frequency for all the nodes.
The experimental results show that the proposed IVS policy can provide up to
29% energy savings and is competitive with more complex schemes for some workloads. VOVO policy can produce saving up to 42%, whereas CVS policy in conjunction with VOVO (VOVO-CVS) results in 18% higher savings that are obtained using VOVO separately. However, the proposed approach is limited in the following factors. The transition time for starting up an additional node is not considered. Only a single application is assumed to be run in the cluster and the load- balancing is supposed to be done by an external system. Moreover, the algorithm is centralized that creates a single point of failure and reduces the system scalability. The workload data are not approximated, which can lead to inefficient decisions due to fluctuations in the demand. No other system resources except for CPU are
considered in resource management decisions.