(GSLA) framework for cloud computing - Springer Link

Computing (2016) 98:949–963 DOI 10.1007/s00607-015-0481-6

Green Service Level Agreement (GSLA) framework for cloud computing Sudhir Goyal1 · Seema Bawa1 · Bhupinder Singh2

Received: 29 October 2014 / Accepted: 21 December 2015 / Published online: 19 January 2016 © Springer-Verlag Wien 2016

Abstract As the organizations are shifting their workload on cloud computing, the demand of cloud services has increased tremendously. With the increased usage of cloud data centers, there is huge consumption of energy (power and heat), contributing to high operational costs and carbon footprints to the environment. So far, research has been carried out to optimize energy usage for cloud resources. However, most of the work on energy optimization is centered on the operational phase of a data center. This paper focuses on energy reduction at Service Level Agreement (SLA) level. Cloud resources are provisioned with Green SLA aware cloud resource reservation (GSLACRR) algorithm. This work proposes Green Service Level Agreement (GSLA) template and negotiation strategies for cloud services. It offers cloud resource services in an energy efficient manner to the users. Keywords Service Level Agreement · Resource provisioning · Energy efficient cloud · Energy-aware VM placement · Virtual machine Mathematics Subject Classification

B

68N01 · 68U01 · 68M01 · 68M14 · 62H15

Sudhir Goyal [email protected] Seema Bawa [email protected] Bhupinder Singh [email protected]

1

Department of Computer Science and Engineering, Thapar University, Patiala, India

2

Microsoft India (R & D) Ltd., Hyderabad, India

123

950

S. Goyal et al.

1 Introduction To minimize the operating and procurement cost of Information Technology (IT) infrastructure, many big and small enterprises are shifting their workload on cloud computing. This leads to a formal legal agreement known as Service Level Agreement (SLA) between cloud provider and service user about the offered Quality of Service (QoS) and the cost model. The SLA for cloud computing is a negotiation between cloud provider and the users under which cloud provider must provide their services within the ambit of QoS parameters such as uptime of service availability, security, cost, service incentives and penalties on violation of agreement. Until now, most of the work on energy optimization has been centered at the data center level. Energy optimization at the data center level is performed through virtual machine (VM) placement, consolidation and migration [1–4]. Through this VMs are consolidated on a subset of servers and switching the rest of the servers into low power mode. Few authors have proposed energy minimization work at server and cluster level using Dynamic Voltage Frequency Scaling (DVFS) technique [5–9]. In DVFS, frequency of CPU is adjusted according to the workload to optimize the power consumption. Another thread of research for energy optimization focused on thermal aware resource management [10,11]. In thermal-aware scheduling, tasks are scheduled to minimize the overall data center’s temperature. The goal of thermal aware scheduling is to reduce the data center-wide thermal gradient, hotspots and cooling magnitude [12]. Energy optimization at SLA level has not been much emphasized in the past work. Public cloud vendors have large IT infrastructure to cater computational need for large number of users. However small private cloud providers have limited resource availability in comparison to public cloud service providers like Amazon, Google, Microsoft, IBM etc. It will be a challenging goal to cater computational need of large number of users. In addition to that addressing the issues of energy efficiency without severe loss of performance make the whole process more challenging. This paper proposes energy aware SLA or Green SLA (GSLA) for a private cloud provider, to offer resource services in an energy efficient manner to cloud users. The contribution of this paper as follows. • GSLA architect has been framed that outlines the general negotiation structure and approach. • A framework is implemented and tested for academic environment like universities. • GSLA aware Cloud Resource Reservation algorithm has been developed to provide resources energy efficiently. The rest of the paper is organized as follows. Section 2 has related work. A Green SLA management framework and energy based resource provisioning policy (EBRPP) has been discussed in Sect. 3. Cloud model, GSLA aware cloud resource reservation algorithm and experiment setup has been explained in Sects. 4 and 5 respectively. Results have been discussed in Sect. 6. Finally, conclusion and future work has been made in Sect. 7.

123

Green Service Level Agreement...

951

2 Related work There is a need to involve users in green computing for environmental, social and economic sustainability. Various authors have proposed SLA based energy conservation for cloud data centers to achieve the same. In the following section, prior work on Green SLAs and renewable energy SLAs have been discussed. 2.1 Green SLAs In [13], the authors have proposed Green SLA for high performance cloud computing providers, in which each client specifies the minimum energy consumed for the submitted job. On violation of Green SLA, the provider is penalized. On the similar research thread, an energy optimization approach has been proposed through GSLA that enhanced text-based SLA by including semantic information regarding metrics and behavior [14]. Energy data has been collected on a per service request basis. In addition, GSLA also offers incentives to users to trade off the traditional performance parameters and matrices. However, this work does not collect energy data per request basis as it has been observed that most of the users are not concerned with the specification of energy budget in advance. Another approach for green SLA was developed by Laszewski and Wang [15], which focused on measuring and monitoring the eco- efficiency of green IT services offered by data center. Green IT-SLA uses energy metrics to monitor the eco-efficiency of the offered services. Contrary to the authors work, the proposed work is more inclined towards offering different energy efficient solutions to users. Users can see which offer is suitable for them. An energy aware VM placement has been presented in [16] using constraint programming. The authors designed an optimizer which was the core element of energy aware cloud framework. It handled SLA requirements and minimizes the energy consumption on federated cloud data centers. The authors proposed framework [16] deals with a federation of different interconnected data centers and each have its own characteristics. However, the proposed work is specifically confined to one data center for small (IaaS) provider. Particle swarm optimization (PSO) based negotiation for SLA has been proposed in [17]. PSO evolves populations of possible negotiation offers for each party about a new offer. This process leads to an approximate negotiation solution. The authors have proposed negotiation work biased towards promoting the decreased resource allocation within the cloud customer’s accepted limits [17]. However, this work inclined users to make a balance between energy and performance. OPTIMIS toolkit project of Rasheed et al. [18], has focused on optimizing usage of Cloud infrastructure services based on the parameters such as trust, risk, eco-efficiency and cost (TREC parameters) specified in SLA. Further, the authors developed and evaluated an approach for service manifesto and SLA on multiple cloud architecture. Basically, trust model has an evaluation process at user level where cloud providers are selected based on high level of trustworthiness. It is more suitable for critical business application and more sensitive information. However, this work does not emphasize on trustworthiness and moreover software running inside the VM are used for application development and non critical activities.

123

952

S. Goyal et al.

Gaun et al. [19] have proposed integrated power management solution for virtualized data centers which takes the benefit of VM resizing and server consolidation for energy conservation and QoS defined in SLA. The proposed framework consists of performance controller and an energy optimizer. The performance controller maintains the demanded performance through dynamic VM resizing. The energy optimizer consolidates VMs onto the power-efficient servers by using ant colony optimization (ACO) algorithm. In contrast, this work takes the user in confidence about the size of VM initially at the time of SLA instead of dynamic VM resizing afterward. In [20], Chen et al. have proposed energy aware SLA (EASLA) scheduling algorithm for precedence-constrained applications. This algorithm minimized the energy consumption within the permissible limits of makespan of tasks. Under the proposed architecture, user negotiates with the service provider about makespan extension of a task to reduce the energy consumption under the QoS. However, the proposed approach does not consider the makespan extension of submitted tasks. FIT4Green project [21] has proposed an energy-aware computing framework. It has comprised of energy saving strategies and planning. These strategies have been packaged and used in the context of data center control frameworks. Similarly in Plug4Green framework [22] dealt with technical SLA and energy related constraints through constraint programming (CP). The proposed solution triggers two events: single allocation and global optimization. Single allocation event triggers on new VM allocation. It maps VM to a server while taking care of VM characteristics, current state of servers, SLA constraints and current objectives of data center in terms of minimizing the power consumption or CO2 emission. Global Optimization event executes itself regularly and produces data center reconfiguration plan (switching on or off a server, migrating a VM etc.) as an output. The projects of authors [21–23] have been designed and implemented with the context of data center control framework. However, the proposed work handles users’ requests tactically and inclined them toward sustainable computing.

2.2 Renewable energy SLAs Renewable energy for a data center means sourcing data center power from renewable sources such as sunlight, wind, rain, tides, waves and geothermal heat. It reduces carbon emissions. Google [23] is using 35 % of power for their operations from renewable energy. Deng et al. [24] have discussed the notion of adaptive cloud hosting where users have been given an opportunity to specify the renewable energy percentage for powering their services. It investigates clean energy cost for existing green hosts. Other recent work focused on improving the energy efficiency of data center through shifting the workloads to data centers having renewable energy sources. Eduard Oró et al. [25] have presented work on renewable energy integration into data centers and analyzed its characterization using numerical models. The authors throw light on present scenarios of data center, its environmental guidelines, business models, power distribution system and cooling system. It has further summarized a number of energy efficiency strategies.

123


953

Chao Li et al. [26] have provided solution for power/carbon constrained data center that needed to scale out with a large number of compute nodes. The authors have proposed a power provisioning scheme, Oasis, that could scale out data center economically and sustainably. In addition to that, a multi-source driven power management scheme, Ozone, has been proposed. The results of their study revealed that Ozone could help Oasis to reduce workload execution delay to 1 % and extend battery lifetime by 50 %. In [27], the work has focused on the usage of onsite distributed generation (DG) sources for providing a clean energy to the computing load. In this work a novel technique, namely power demand shaping (PDS), has been proposed that allows data centers to utilize onsite green energy efficiently. From the performance evaluation, it has been concluded that their technique gives 1.2 % better performance than an ideal oracle which has roughly a 37 % improvement over existing supply tracking based design. Deng et al. [28] have emphasized on the usage of a grid-tie device that integrate renewable energy into the data center power delivery system. A renewable-powered instances as a metric has been proposed that measures the concentration of renewable energy in data centers. Through the simulation, the authors have found that grid-tie placement has effects on renewable-energy concentration and it reduces the use of grid energy. The above discussed study urges users to value renewable-powered computational hosts. It emphasis to incorporate renewable energy percentage into users’ contracts and in lieu of that users get incentives to use more resources if they are powered by sustainable energy sources. Other thread of research on energy optimization was focused on VM resizing, placement, migration and (re-)configuration to make energy optimized operations at data centers level. In this work, efforts have been put up to make a balance between performance and energy conservation through Green SLA. In the proposed work, a private cloud has been setup in the university premises which is used for academic workload. Workload is almost predictable and stable as most of the work activities are performed as per pre-defined schedule. Proposed Green SLA is designed to make resource usage policy eco-friendly.

3 Green Service Level Agreement (GSLA) management framework GSLA template (Fig. 1a) is an extended version of traditional SLA by introducing energy optimization offer. The framework (Fig. 1b) consists of cloud service provider and user’s negotiation and document generation. It consists of following steps. Creation GSLA creation takes input details of user, metric values for every SLA parameter, hardware, software and time slot requirement. Negotiation Negotiation offer deals with various time slots, cost, awards, penalties, service terms, energy optimization incentives and alternative hardware configuration to users with requisite software. Implementation After negotiation, request can be denied or resources can be reserved with time slot for user id. Apart from the above, there is a negotiation on minimum energy consumption with the following points.

123

954

S. Goyal et al.

GSLA offer id User Name GSLA Energy minimization Offer SLA Parameters Awards, Penalties & Cost

Service Terms

(a)

(b)

Fig. 1 Green Service Level Agreement

3.1 Energy based resource provisioning policy (EBRPP): a negotiation approach EBRPP has been designed on the basis of requested resources, time slot and energy consumption parameters. When a user submits a request, resource manager checks the availability of resources, if available, a further negotiation is dialogued on the SLA parameter metric values and energy minimization policy as discussed below. In case of non-availability of resources, a possible time slot for resource availability is negotiated. The main objective of EBRPP is to provide the resources and minimize the energy consumption. A negotiation takes place on the following points. 1. Resource reservation: As the resource reservation request received from the user, resources are reserved with demand time slot on the availability of resources. If the scheduler cannot make the reservation at that time, users will have two options; (a) the time slot which is near to user’s requested time slot will be allocated to him or (b) he is granted time slot on weekend. The objective of negotiation is to aggregate all the reservations into one part of slot from big free time slot and switch off resources into other part of the time slot [29]. 2. Credits: Credits are added to the corresponding user account that agree with resource usage time slot and performance modifications through negotiation. If user agrees to work on specified time slot, user gets 1 % credits in his/her account. 3. VM running mode: Generally specified resources for VM are overestimated, based on the assumption for peak workload. Moreover, some tasks use resources and after a while it remains idle and still consumes power. So, there is negotiation between cloud service provider and cloud user about the sleeping time of VM when it remains idle. When there is no interaction with the VM for a specified number of minutes, VM will be automatically switched to sleep mode. In the same way, it will take at least some amount of time to reactivate when user again interacts with the VM.

123


955

4. Power budget policy: cloud provider offers a power budget policy to maximum usage of power by the requesting user. When user used resources reaches to some power limit like (80 %), a warning message will be generated to the user so that he can manage his work according to that. This can be optional, as the most of the users are not concerned with the specification of power budget in advance. Apart from the above, some of the standard performance parameters included are as follows 1. Availability of data: the requested data should be available 99 % of the times. 2. Uptime of resources: specifies the uptime of the VM with special provision of point 3 in above mentioned negotiation points. 3. Disaster recovery: in the event of untoward incident, the mean time for the recovery of the resource services. 4. Elasticity: it specifies the size of a resource can grow. 5. Cost: it specifies the cost of offered services with respect to time and resources usage. Also, other SLA parameters like security, privacy, location of data, portability of data, legal issues, dispute process, exit strategies etc. can be included. It all depends upon the criticality of data and applications. As private cloud is deployed inside the premises of organization, so these parameters are not considered in the proposed framework.

4 Case study: SLA for academic sphere To demonstrate the effectiveness, a framework is implemented and tested for academic environment. The proposed framework uses a credit system. That credit can be exchanged with money in public commercial cloud system. Each student gets 100 credits in advance in its account on the start of semester. Cloud instance credits are calculated based on the offering resource sizes. For using large VM computational services for 50 h, 5 credits are deducted. Credits add to student’s account based on the failure to meet any specific part of GSLA and the unavailability percentage of the instance out of 167 h on an average per month. Student account will be credited 1 % of the monthly usage credits for every hour of downtime. 4.1 System model This section describes the cloud model integrated with Green SLA (Fig. 2) which provides the cloud resources as per agreement. In an academic sphere, the resource requests come either from teacher or student/researchers. Each request comprises of a number of VMs, VM type (small: 1 VCPU, 1 GB RAM, 40 GB HDD; medium: 2 VCPU, 2 GB RAM, 100 GB HDD; large: 4 VCPU, 4 GB RAM, 200 GB HDD), operating system, execution end time. 4.1.1 Reservation model As shown in Fig. 2, VMs requests are submitted by the following procedure.

123

956

S. Goyal et al.

Fig. 2 Cloud system model

1. Resource allocation queue (RAQ) takes all VMs requests submitted by a teacher. VMs assigned in RAQ must be provisioned on the PMs without any negotiation as these are high priority jobs. 2. Student/researchers VMs request is negotiated under GSLA. if GSLA status is accepted, requested VMs assigned into resource waiting queue (RWQ) and communicated to user under GSLA negotiation status information. On request rejection, negotiation complete (Status: Rejected) message communicated to user. 4.1.2 GSLA aware cloud resource reservation (GSLACRR) algorithm The acronyms used in the algorithms are listed in Table 1. GSLACRR, Algorithm 1, takes input of VMs request from teachers and students. Steps 2 and 3 processes the teacher’s VM request and stored in the RAQ[] as discussed in Sect. 4.1.1. Steps 5 to 25 deals with VM requests of students and stored in the RWQ[] on successful negotiation. Step 15 calls energy optimized VM placement Algorithm 1.1 and handles present workload. Algorithm 1.2 [30] migrates the VM among the set of PMs; takes care the overhead of VM migration and switches active PM into sleep state

123


957

Table 1 Notation and their meanings used in algorithms Symbol

Meaning

PMactive

A set of active physical machines/servers

PMinactive

A set of inactive/sleeping physical machines

RAQ[]

Resource allocation queues (RAQ) takes VMs request submitted by a teacher according to students’ strength and type of applications required to students for conducting laboratory class

RWQ[]

Resource waiting queue (RWQ) takes VMs request submitted by students/researchers for their research work

pmi.inactivit ytime

Physical machines i’s inactivity time is a time during which physical machine remains idle. In laboratory time table slot, there could be a free times slot during which PM remains idle/low workload

pmi.therasoldtime

Physical machines i’s switching threshold time is the time interval when energy consumption of two possible cases are equal a) the resource is switched off, stays off for a while and then switches it on again, b) the case where the resource stays idle for the entire interval

VM j.remainingcompletiontime

VM’s remaining completion time is the time left to complete the VMs execution from the present time. it is VM j.duration (VM j. pr esenttime - VM j.star tedtime )

VM j.migrationtime

Time taken to migrate a VM from one physical machines to another

123

958

S. Goyal et al.

5 Performance evaluation In this work, GSLACRR algorithm is compared with non GSLA aware algorithm. In a non GSLA aware algorithm, there is simple negotiation between user and cloud provider. On user’s request, if resources are available for the requested time slot, then it is allocated otherwise a reject message is conveyed to the user. Also in non GSLA aware, there is no ranking of servers and order of VM specified. In order to compare the efficiency of the proposed GSLACRR algorithm, two performance metrics have been used. First performance metric is the mean energy consumption (MEC) in which mean energy consumption of all computational resources for the predefined schedule intervals are compared. It measured in kWh. Second metric is accepted VM (AVM)

123


959

Table 2 Energy consumption parameters of the physical machines [33–36] Server type

Number

Cores (mi )

Model

Pidle (W)

Pimax (W)

Dell power edge r710 server

3

2*4

Xeon E5620

128.7

247

Dell PowerEdge 2900 server

1

2*4

Xeon®5400

40

80

Dell i5 2.5 GHZ

1

4

i5 2.5 GHZ

39.8

106.8

Dell Core 2 Duo E7500

2

2

Core 2 Duo E7500

39.3

78.9

that measures the number of accepted VM requests out of total given requests for all scheduled intervals. 5.1 Experimental setup For experiment, a heterogeneous private cloud has been setup, named Academic Cloud (ACA-Cloud), consisting of seven machines with data as given in Table 2. All machines are connected with 100 Mbps Ethernet and 8 TB shared SAN-based storage. As a virtual machine manager, Kernel-based virtual machine (KVM) [31] is installed on each machine. Hosts run virtual machines and a common storage provides seamless storage to allow live migration of VMs. All the decisions regarding the VM migration, starting and shutting down of cloud nodes are taken by ACA-Cloud resource scheduler. As a base operating system, ubuntu 12.4 [32] is installed on each machine. 5.2 Workload data For experimental results, realistic workload of cloud has been taken. Workload trace log has been generated based on the software used for academic sphere run inside VM on a private cloud. There are three standard VMs allocated to users (small: 1 VCPU, 1 GB RAM, 40 GB HDD; medium: 2 VCPU, 2 GB RAM, 100 GB HDD; large: 4 VCPU, 4 GB RAM, 200 GB HDD). Small VMs are allocated to users who run simple program of C, C++, Java, Oracle etc. Netbeans, eclipse, IIS web server, SPSS run on medium sized VM. On large fedora VMs, physics experiments have been executed like physics scholars use statistical multi fragmentation model (SMM), quantum molecular dynamics (QMD), isospin dependent quantum molecular dynamics (IQMD), Boltzmann-Uehling-Uhlenbeck (BUU). Power consumption reading has been taken after each minute.

6 Results and discussion Figure 3, presents comparison chart of GSLACRR and non GSLA aware algorithms for a number of accepted VMs request against the total number of the VM requests submitted. Figure 4, depicts the average power consumption of all the PMs against the total number of the VMs requests submitted. Power consumption of a PM is basically dependent on a number of VM running, VM’s size, duration and numbers of PMs

123

S. Goyal et al. 90 80 70 60 50 40 30 20 10 0

Schd. 27,83

Schd. 29,87

Schd. 31,91

Schd. 33,95

Schd. 35,99

Schd. 37,103

Schd. 39,107

Schd. 27,83

Schd. 29,87

Schd. 31,91

Schd. 33,95

Schd. 35,99

Schd. 37,103

Schd. 39,107

Schd. 25,79

Schd. 23,75

Schd. 21,71

Schd. 19,67

Schd. 17,63

Schd. 15,59

Schd. 13,55

Schd. 11,51

Schd. 9,47

Schd. 7,43

Schd. 5,39

Schd. 3,35

GSLACRR Non GSLA Aware

Schd. 1,31

Accepted Requests

960

Schedules and Total Reqests

Schd. 25,79

Schd. 23,75

Schd. 21,71

Schd. 19,67

Schd. 17,63

Schd. 15,59

Schd. 13,55

Schd. 11,51

Non GSLA Aware

Schd. 7,43

Schd. 5,39

Schd. 3,35

GSLA Aware

Schd. 9,47

100 90 80 70 60 50 40 30 20 10 0

Schd. 1,31

Power Consumption (kWh)

Fig. 3 Accepted VM request

Schedules and Total Request

Fig. 4 Avg. power consumption of all physical machines

in sleeping state kept by the algorithm. There is a trade off accepted VMs requests and energy conservation. When the number of VMs requests are less in proportion to available infrastructure, the proposed algorithm saves power by consolidating VMs on fewer physical machines as well as takes care of VM migration overhead. As the number of users requests increases, a non GSLA aware algorithm need to active all the physical machines and rejects new VMs requests beyond the available infrastructure capacity. The proposed algorithm also addresses the increased VMs requests tactically. The results reveal that GSLACRR aware algorithm performs better in terms of acceptance of VM requests as well as power consumption. The analysis reveals that proposed GSLACRR algorithm performs much better in terms of power consumption up to schedule22 when number of VM requests are equal as shown in Figs. 3 and 4 respectively. After schedule22 in Fig. 4, GSLACRR algorithm consumes more power but it accepts more requests for the corresponding schedules as shown in Fig. 3. GSLACRR aware algorithm performs better with respect to MEC on receiving a less number of VM requests or having small VM types. As mentioned in the algorithm, servers are ranked with the ascending order of VCPUs. When a number of VM requests

123


961

Table 3 t-Test: paired two sample for means for MEC metric GSLACRR

Non GSLA aware

Mean

67.91966834

68.95377817

Variance

356.854887

293.9093177

Observations

40

40

Pearson correlation

0.989967185

Hypothesized mean difference

0

df

39

t Stat

−2.116412915

P(T ≤ t) one-tail

0.020372278

t Critical one-tail

1.684875122

P(T ≤ t) two-tail

0.040744555

t Critical two-tail

2.022690901

are lower, most of VM requests are accommodated by high power efficient and large core servers. Remaining servers are in low power mode. Secondly, through negotiation all the VM requests reservations are aggregated into one part of slot from big free time slot and switch off resources into other part of the time slot. In non GSLA aware, there is no specific ranking of servers and the number of active servers were more all the time in comparison to GSLACRR algorithm. On receiving large number of VM requests, GSLACRR algorithm accepts more VM requests because large sized VM requests are accommodated by high power efficient and large core servers. Medium and Small VM requests are allocated to small core servers. With this implementation, GSLACRR algorithm is able to accommodate more VM requests. Now in non GSLA algorithm, there is no ranking of servers and order of VM specified. So medium and small VMs requests are allocated to large core servers leading to their unavailability at the time when a large VM requests are arrived. For statistical validation, t-Test has been conducted where the MEC of GSLACRR algorithm is compared with the MEC of another non GSLA aware algorithm. The null hypothesis is that the MEC of two algorithms are same. The results of MEC of GSLACRR algorithm with non GSLA aware algorithm are reported in Table 3. The results have been found significant at 0.05 levels. The results of GSLACRR algorithm with non GSLA aware for AVM metric are reported in Table 4. t-Test results in Tables 3 and 4 clearly indicate that proposed algorithm is significantly different from the non GSLA aware algorithm with regard to the MEC and AVM metrics.

7 Conclusion In this paper, Green SLA management and an EBRPP have been discussed for a small IaaS cloud. The aim is to minimize the energy consumption of a data center. Green SLA devises the provision of resources through negotiation with user. The proposed GSLACRR algorithm leverages the free time table slots and free weekend days, and adjusts users request in these slots through negotiation to enhance the user acceptance

123

962

S. Goyal et al.

Table 4 t-Test: paired two sample for means for AVM metric GSLACRR

Non GSLA aware

Mean

62.2

61.675

Variance

224.7282051

210.7891

Observations

40

40

Pearson correlation

0.998864712

Hypothesized mean difference

0

df

39

t Stat

3.920455738

P(T ≤ t) one-tail

0.000173559

t Critical one-tail

1.684875122

P(T ≤ t) two-tail

0.000347118

t Critical two-tail

2.022690901

rate. This work has evaluated the proposed GSLACRR algorithm with non GSLA aware. The results show that GSLACRR significantly improves acceptance rate as well as energy consumption by user negotiation. In the future, more complicated workload cases will be analyzed and make this approach generalized.

References 1. Beloglazov A, Abawajy J, Buyya R (2012) Energy-aware resource allocation heuristics for efficient management of data centers for cloud computing. Futur Gener Comput Syst 28(5):755–768 2. Minh QD, Federico M, Domenico S, Giafreda R (2012) T-Alloc A practical energy efficient resource allocation algorithm for traditional data centers. Futur Gener Comput Syst 28(5):791–800 3. Lee YC, Zomaya AY (2012) Energy efficient utilization of resources in cloud computing systems. J Supercomput 60(2):268–280 4. Beloglazov A, Buyya R (2012) Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in cloud data centers. Concurr Comput Pract Exp 24(13):1397–1420 5. Rizvandi NB, Taheri J, Zomaya AY (2011) Some observations on optimal frequency selection in DVFS-based energy consumption minimization. J Parallel Distrib Comput 71(8):1154–1164 6. Beloglazov A, Buyya R (2010) Energy efficient resource management in virtualized cloud data centers. In: Proceedings of the 2010 10th IEEE/ACM international conference on cluster, cloud and grid computing, pp 826–831 7. Kim KH, Beloglazov A, Buyya R (2009) Power-aware provisioning of cloud resources for real-time services. In: Proceedings of the 7th international workshop on middleware for grids, clouds and escience, p 1 8. Kan EY, Chan WK, Tse TH (2012) EClass: An execution classification approach to improving the energy-efficiency of software via machine learning. J Syst Softw 85(4):960–973 9. Guzek M, Diaz CO, Pecero JE, Bouvry P, Zomaya AY (2012) Impact of Voltage Levels Number for Energy-aware Bi-objective DAG Scheduling for Multi-processors Systems. Advances in Information Technology. Springer, Berlin, pp 70–80 10. Sharma RK, Bash CE, Patel CD, Friedrich RJ, Chase JS (2005) Balance of power: Dynamic thermal management for internet data centers. Internet Comput IEEE 9(1):42–49 11. Moore JD, Chase JS, Ranganathan P, Sharma RK (2005) Making Scheduling Cool: TemperatureAware Workload Placement in Data Centers. In USENIX annual technical conference, General Track, pp 61–75

123


963

12. Chaudhry MT, Ling TC, Manzoor A, Hussain SA, Kim J (2015) Thermal-aware scheduling in green data centers. ACM Comput Surv (CSUR) 47(3):39 13. Haque ME, Le K, Goiri Í, Bianchini R, Nguyen TD (2013) Providing Green SLAs in High Performance Computing Clouds. In: IEEE international Green computing conference (IGCC). IEEE, Arlington, pp 1–11 14. Bunse C, Klingert S, Schulze T (2012) GreenSLAs: Supporting energy-efficiency through contracts. Energy Efficient Data Centers. Springer, Berlin, pp 54–68 15. von Laszewski G, Wang L (2010) GreenIT service level agreements. In: Grids and Service-Oriented Architectures for Service Level Agreements Springer, US, pp 77–88 16. Dupont C, Giuliani G, Hermenier F, Schulze T, Somov A (2012) An energy aware framework for virtual machine placement in cloud federated data centres. In: IEEE third international conference on future energy systems: where energy, computing and communication meet (e-Energy), pp 1–10 17. Copil G, Moldovan D, Salomie I, Cioara T, Anghel I, Borza D (2012) Cloud SLA negotiation for energy saving—a particle swarm optimization approach. In: IEEE International Conference on Intelligent Computer Communication and Processing (ICCP), pp 289–296 18. Rasheed H, Rumpl A, Wäldrich O, Ziegler W (2012) A standards-based approach for negotiating service QoS with cloud infrastructure providers. In: eChallenges Conference 19. Gao Y, Guan H, Qi Z, Wang B, Liu L (2013) Quality of service aware power management for virtualized data centers. J Syst Arch 59(4):245–259 20. Chen X, Li K, Liu C, Li K (2014) SLA-based energy aware scheduling of precedence-constrained applications on DVFS-enabled clusters. In: 20th IEEE international conference on parallel and distributed systems (ICPADS). IEEE, Hsinchu 21. Basmadjian R, Bunse C, Georgiadou V, Giuliani G, Klingert S, Lovasz G, Majanen M (2010) Fit4greenenergy aware ICT optimization policies. In: Proceedings of the COST Action IC0804 on energy efficiency in large scale distributed systems—1st year, pp 88–92 22. Dupont Corentin et al (2015) Plug4Green: a flexible energy-aware VM manager to fit data centre particularities. Ad Hoc Netw 25:505–519 23. http://www.google.com/about/datacenters/renewable/. Accessed 7 Jan 2016 24. Deng N, Stewart C, Gmach D, Arlitt M, Kelley J (2012) Adaptive green hosting. In: ACM Proceedings of the 9th international conference on Autonomic computing, pp 135–144 25. Oró E, Depoorter V, Garcia A, Salom J (2015) Energy efficiency and renewable energy integration in data centres. Strategies and modelling review. Renew Sustain Energy Rev 42:429–445. doi:10.1016/j. rser.2014.10.035 26. Li C, Hu Y, Zhou R, Liu M, Liu L, Yuan J, Li T (2013) Enabling datacenter servers to scale out economically and sustainably. In: Proceedings of the 46th annual IEEE/ACM international symposium on microarchitecture, pp 322–333 27. Li C, Zhou R, Li T (2013) Enabling distributed generation powered sustainable high-performance data center. In: IEEE proceeding of the 19th international symposium on high performance computer architecture (HPCA2013), pp 35-46 28. Deng N, Stewart C, Li J (2011) Concentrating renewable energy in grid-tied datacenters. In: IEEE proceeding of international symposium on sustainable systems and technology (ISSST), pp 1–6 29. Orgerie AC (2011) An energy-efficient reservation framework for large-scale distributed systems. PhD thesis, Ecole Normale Supérieure de Lyon–France 30. Goyal S, Bawa S, Singh B (2015) Energy optimized resource scheduling algorithm for private cloud computing. International Journal of AdHoc and Ubiquitous Computing, Inderscience (In Press, Accepted Manuscript) 31. Kernel-based Virtual Machine: http://www.linux-kvm.org/page/Main_Page 32. Ubuntu operating system. http://www.ubuntu.com/ 33. Power consumption of Dell PowerEdge r710 server (2014) http://www.dell.com/downloads/global/ products/pedge/en/dell_poweredge_r710_2p_e5620_870w_energy_star_data_sheet.pdf 34. Power consumption of Dell PowerEdge 2900 server (2014) http://www.intel.com/content/dam/www/ public/us/en/documents/datasheets/quad-core-xeon-5400-datasheet.pdf 35. Gavrichenkov I, CPU Benchmark (2014) http://www.xbitlabs.com/articles/cpu/display/ cpu-benchmark-mainstream_10.html 36. Gavrichenkov I (2014) Power consumption of intelCore i5 processor. http://www.xbitlabs.com/articles/ cpu/display/core-i5-2500-2400-2300_10.html

123