A Critical Analysis of Green Database System for Energy Efficiency and Green Computing

(1)

A Critical Analysis of Green Database System for

Energy Efficiency and Green Computing

Kanahavalli Mardamutu, Ryan Anthony Lim Aik Joon, Jayanand Jegatheesan, Vasaki Ponnusamy and Ying Mei

Leong

Faculty of Integrated Science and Technology, Quest International University of Perak, QIUP, Ipoh, Perak Darul Ridzuan, Malaysia

{sf00220, sf00238, sf00122, vasaki.ponnusamy, [email protected]}

Abstract – Energy efficiency is highly significant in today’s high-performance data center and storage systems. Decades of research have explored factors that influence the increasing usage of energy. That conventionally improves performance like single-rotation-rate, performance, throughput, slack time reduction in data storage systems. However, the influence of physical characteristics of energy-saving measures on overall acceptability is largely ignored by the users and service providers. The need to reduce greenhouse gas effect calls for energy-efficient technologies through green computing. This paper embarks into the study on preferences and analysis done for different types of energy-saving measures and green database management system. Most of the existing solutions compromise the performance for energy conservation, Green Database (GreenDB) and green computing propagation. A comprehensive critical analysis on energy efficiency and GreenDB was carried out and a redundancy-based solution was proposed to address this problem. This work differs from others in the literature by evaluating the essence of solution provided by many groups of research. Precise overview of analysis projected to visibly prove the cream of each research, thus proposing a wholesome solution for energy efficiency in green computing.

Keywords-Database, Energy-saving, Energy-efficiency, Power-saving, GreenDB, Green Data Center, Data Center, Cloud Computing, Green Database

I. INTRODUCTION

The increase in need of data processing has led to a demand in obtainability for cheaper, faster, efficient and larger data management system. This has further introduced a significant problem in terms of energy consumption as mentioned in [1]. This has further lead to a significant problem in terms of energy usage [1]. Since the trend of Internet Infrastructure is moving towards services-based computing, data center plays a key role in this new computing architecture. It is used widely in variety of services including web hosting, application services, electronic markets, outsourced storage, online-monitoring system and other network services. Environment Protection Agency shows that the data center and database management system in the US consume power doubled between 2000 and 2006 and may multiply again within the next decade [2].

Many works have been done to produce efficient result in reducing the energy usage for the purpose of data center and

data management for example by Cisco System Inc., Google Inc., and many more [1]. Some of the approaches were to design highly efficient systems at the peak performance point with maximum utilization of resources. But very few studies have addressed the energy problem comprehensively [3].

II. PROBLEM STATEMENT

In today‟s technological world, it is proven that there is a significant relationship between data center and investment in capital overlay and outgoing costs. Table 1 shows the detailed breakdown on the distribution of the cost from data center [22]. The design of computing system is being optimized based on the execution time and operates at low consumption due to entrapment of resources and fragmentation of data. No concern was made for the issue of energy consumption by the system. Uncontrolled usage of energy in data centers has negative effects on the reliability, density, scalability of information processing and the environment. Need and awareness are raised across nation through multiple solution and disciplines to optimize and save energy use in data management system [1, 2, 3]. System or even its components like CPUs, memory and disk are hardly distributing energy efficiently. It trades power for performance [2]. Power management has become a critical issue due to large energy usage and all over the world government has started imposing the taxes on the carbon emission, ultimately aiming to reduce the emission and help make computing into green computing [12].

Although research was conducted in designing power aware applications, an effort for implementing has not been taken place. Large expenses and allocations need to be augmented when data center reaches maximum provisioned power [11]. In the very near future, energy eﬃciency is expected to be one of the key procuring arguments in the society. Research should focus in building power-aware database management systems (DBMS) or GreenDB.

(2)

TABLE I. GUIDE ON THE DISTRIBUTION OF DATA CENTER COSTS.

Amortized Cost Component Sub-component

~ 45% Servers CPU, memory, storage systems. ~ 25% Infrastructure Power distribution and cooling. ~15% Power draw Electrical utility costs. ~15% Network Links, transit, equipment.

a. Source of data: A. Greenberg et al, “The cost of a cloud: research problems in data center networks”, ACM SIGCOMM Computer Communication Review, 39(1), pp. 68, 2008

III. MOTIVATION

Analysis shows that giant companies are also working towards energy efficiency using green database system. Huge volume of energy usage goes to the data center.

The global business market intelligence is a billion dollar market and still growing to see double-digit growth rates [3]. With the growth of the internet data center (IDC), data systems are also running its management workload such as search engines, storage of multimedia like video and audio, streaming, web analytics, and cloud computing. Dataset are growing rapidly and workload is also increasing. Together with it, the energy usage has increased. Companies like Google are working towards reducing the energy usage excessively in data centers. [3]. Data center has become the major immobilizer in the topic of energy wastage.

Clients particularly are searching for high scalability setup, lower unit price and simplicity of management. As to maintenance costs, the second largest bill for large database centers comes from the power company. Statistic shows that in US alone, data centers consumed about 60 billion kilowatt-hours of electricity in 2006 which equaling to 1.5% of electricity consumption in US [2]. Demand for a solution is in the high urge and is being researched by many giant companies.

IV. RESEARCH AIMS

The uniqueness of our work and our aims are as follows:

 To investigate energy-aware research conducted on database management in a way that improves energy efficiency of a data center.

 To explore possible solution that may be achievable through green computing architecture.

V. RESEARCH QUESTIONS

The above mentioned research aims have led to address the following research questions:

 What are the power consumption patterns in energy-aware DBMS?

 What are the power-saving opportunities in GreenDB?

 How power consumption patterns in GreenDB

conserved energy?

 How power-saving databases are being optimized?

 What are the measures in employing energy efficiency via green computing and green database system?

 How these determinants could be best implemented for energy efficiency and green computing?

VI. OBJECTIVES

This research is not to reinvent the wheel to conduct a critical analysis on the existing research to integrate the idea proposed to produce wholesome elucidation. Modules from many different researches have been integrated and evaluated to enhance its components to be fully functional with fewer flaws. Therefore, it is a better solution to show the benefits of energy efficiency through green database. The main objectives of this research are as follows;

1) To initiate and analyse energy-aware research conducted on database management in a way that improves energy efficiency of a data center.

2) To propose solution in implementing the energy efficiency through green computing and green database system.

VII. CRITICALANALYSIS

Many companies, organisations and individuals have proposed methodology to resolve this problem based on the categories shown in Figure 1.

Figure 1. Critical Analysis Flow Model

This research paper analyzes similar papers which has solution but needs minor integration to propose new innovation with better and efficient energy saving. Analysis focuses on power consumption patterns and identification of power-saving opportunity in GreenDB. Critical analysis done based on the following observation.

 Power-aware database.

 Data management through software and hardware.

 Data management through disk power management.

 Green computing architecture

(3)

A. Power-aware Database

Power-aware computing research at the application level in the aspect of hardware and operating system level found to be synergistic for energy proficiency in the core systems. Zinchen Xu [7] describes on creating a power-aware database management (P-DBMS) and initial ideas on the design as such systems, with the focus on a power-aware query optimization module in the DBMS. This particular paper has proposed and created a power model to accurately measure the energy costs of query executions and plan selection. It is in contrast with the heuristic-based adaptive solutions that rely on manual timing and extensive empirical evaluation. It is meaningless without considering the database performance. It is either to make the query run faster by optimizing the query execution or let the system run in lower power state while sacrificing the performance to some extent. They proposed that the throughput is improved while maintaining the power line. It indirectly reduces the power cost while not sacrificing the throughput performance. Implementation has been taken place through initial design in an algorithm based software named PostgreSQL rather than power-saving through hardware and operating system levels modification. The proposed solution resolves the power aware of the database as seen in Table II. Only setback would be database management system needs to be revisited or redesigned and initial efforts be invested. The plan evaluation model has a stumbling block where the system is stable and foreseeable but the database is barely stable and predictable. Thus, modeling error occurs and estimation per operation cost cannot be 100% correct.

TABLE II. CRITICAL OVERVIEW ON POSTGRESQLSYSTEM

Attribute PostgreSQL system

Description Analysis Hardware Hardware Optimisation x Power Management x Software Software Optimisation /

Query management and optimisation /

Architecture

Architecture Manuveuring x

Distributed System x

a. Evaluation based on the result of proposed research paper by Author

B. Data Management through Software and Hardware

Synergy between database management through algorithm and hardware architecture are promising in energy efficiency computing. Stavros Harizopoulus [6] recommends that adjusting existing system-wide configuration and design algorithm for energy proficiency is not the same as designing algorithm based on the throughput and performance. Deploying an energy-efficient data management platform is by choosing hardware components with good performance-per-watt characteristics by choosing high-performance components (server-grade CPUs, solid-state disks etc.). Installing an energy-efficient system was also done also through software-based approaches for reducing wasted energy in data management systems. This paper proposes to power down the

unused hardware components, revisit and redesign data structures and algorithm from a wider perspective. The „push-to‟ design components offers more control over power performance trade-offs for example usage of multicore chips which can be activated at any time. The increase of hardware heterogeneity offers technology refresh cycle like the virtual machine. The disadvantage here is that some workloads may be able to use these additional resources while others will underutilize them and create a waste of power. Table III shows that hardware configuration and application-agnostic power management is only part of the solution proposed in this paper.

TABLE III. CRITICAL OVERVIEW ON STAVROS REVIEW

Attribute Synergistic system

Description Analysis Hardware Hardware Optimisation / Power Management x Software Software Optimisation /

Query management and optimisation /

Architecture

C. Data Management through Disk Power Management

Reducing energy consumption through hardware depends on many aspects. Earlier research shows that the result from monitoring data center is that the average idle period for a server disk is very small compared to the time taken in spinning up and down. This significantly limits the conservation of energy. Qingbo Zhu [5] proposed reducing energy consumption of disk storage using power-aware cache management with algorithm which provides more opportunities for causal disk power management structures in energy efficiency. So far only a few studies [13, 14, 15, 16] addressed the problem in energy efficiency for the storage systems at the data centers. Based on these studies, multi-speed disks or multiple rotational speed disks shown promising results by using synthetics workloads. Therefore, power-aware cache management algorithm was designed whereby it is aware of the base disk energy ingestion and power management plan for read and writes accesses.

It is an energy-optimal cache replacement algorithm which uses the dynamic programming in polynomial time leading to 16% less energy and lead to 50% better average response time. But this study has its limitation such as it is only designed for multiple disks and focuses on high-end storage systems. Single disk may not work with this algorithm. Also it has low effect on prefetching as well. It does not consider the storage cache energy consumption and focuses on smaller part of the overall management as shown in Table IV.

(4)

TABLE IV. CRITICAL OVERVIEW ON POWER-AWARE CACHE

MANAGEMENT ALGORITHM

Attribute Synergistic system

Description Analysis Hardware Hardware Optimisation x Power Management / Software Software Optimisation /

Query management and optimisation x

Architecture

D. Green Computing Architecture

Virtual machine (VM) has been globally used in the data center environment due to its important feature and architecture. Liang Liu [9] uses the latest technology trend which is the cloud computing as a guaranteed performance and designed to reduce the data center energy consumption. It is effective and enables the widespread online-monitoring, live VM migrating and placement optimization. It utilizes the live migration feature of VM Technology such as XEN [12] to implement the architecture for example Amazon Elastic Compute Cloud [17] provides a virtualized computing environment hosted with different kind of Linux-based service. Evaluation results show that 27% energy can be saved when applying Green computing architecture. VM Migration allows lesser usage of computer. This architecture assures real-time performance requirement as well as saves the consumption of energy by Internet Data Center (IDC). Green computing architecture with the integration of VM technology is an effective method in reducing server power consumption.

The problem may arise in the angle of automatic scheduling decision on dynamic migration or consolidating the VMS among physical servers to meet the workload requirements meanwhile saving energy, especially for performance sensitive. The aspect of performance trade-off between the operating system and applications or middleware has not been tested which can be a flaw. But based on Table V, this technology alone may not be sufficient to provide efficiency.

TABLE V. CRITICAL OVERVIEW ON GREEN COMPUTING AND VM TECHNOLOGY

Attribute Green Computing and VM Technology System

Description Analysis Hardware Hardware Optimisation / Power Management / Software Software Optimisation /

Architecture

Architecture Manuveuring /

E. Renewable Energy

Exploiting green energy sources is another opportunity but it is variable and intermittent. So far, there has been less work on integrating renewable energy with database system. Cheng Chen [8] uses the hardware optimization to save energy and retaining the execution part of the hardware. They proposed through integration of renewable energy which is the solar to power the execution of database management system. This research bridges the gap between renewable energy and database management system with ReinDB (Renewable Energy Integrated Database). The purpose would be to minimize the usage on carbon intensive energy consumption on a database server with both renewable energy and carbon intensive energy supplies. This works on Supply Driven Execution (SDE) on a database server with two mechanisms.

a) Workload scheduling to meet workload deadlines

b) Dynamic configurations on hardware setting

c) Brown-aware workload shifting

d) Battery-aware workload shifting

Although it increases the usage of renewable energy (solar usage) but there are still some disadvantages to take into consideration. Studies show that the mismatch between workloads and green energy supplies challenges in the utilization of energy. Divergence in the context of data center from brown energy and green causes excessive usage of energy [18, 19, 20, 21]. Clients are required to change the hardware to new system which may need initial investment that may cost high. Also the latency time (slack time) is also another issue whereby the clients have to tolerate a large slack time to complete a task with the green techniques. In the economical aspect, the cost involves in manufacturing solar panel and battery provision. Table VI, shows that, there are many other aspect to be incorporated.

TABLE VI. CRITICAL OVERVIEW ON INTEGRATION OF RENEWABLE

ENERGY

Attribute Integration of Renewable Energy

Description Analysis Hardware Hardware Optimisation / Power Management x Software Software Optimisation /

Architecture

Architecture Manuveuring /

VIII. RESULT OF ANALYSIS

The concept of green database and energy efficiency is emerging high and gaining more popularity and everyone wants to be on top of it and they have started researches on the concept of energy saving in the DBMSs. Companies like IBM, Oracle that develops their own DBMSs want to decrease the energy consumption. All researches are focusing on individual

(5)

area only. Integration of success methods from each research can result in wholesome solution.

Evaluation and analysis done to multiple proposal and project paper resulting in integration of the best modules for many different sources tosses excellent solution. Comparisons shown in Table VII, VIII, IX, X and XI define the strength and weakness based on the critical analysis done.

TABLE VII. PHYSICAL USAGE

Project/ Attribute Hardware

Modification Software Modification Solar Integration Zichen Xu No No No

Cheng Chen Yes Yes Yes

Liang Liu No Yes No

Stavros Harizopoulos Yes Yes No

Qingbo Zhu No Yes No

TABLE VIII. POWER COVERAGE

Project/ Attribute Power State Power Cost Power

efficiency

Zichen Xu Low Reduced 30%

Cheng Chen Very low Reduced 28-53%

Liang Liu Low Reduced 27%

Stavros Harizopoulos Low Reduced 45%

Qingbo Zhu Low Reduced 16-35%

TABLE IX. QUERY AND THROUGHPUT

Project/Attributes Query Execution Speed Query Optimizer Throughput time

Zichen Xu Faster Increased Maintain

Cheng Chen No change Scheduled Query

peak performance Liang Liu No change Optimized Guaranteed

real time Stavros Harizopoulos No change Optimized Peak

performance

Qingbo Zhu Faster Optimized Limited

TABLE X. OTHER ASPECTS

Project/Attributes Scalability Reliability Slack time Zichen Xu Inefficient Inefficient Retained

Cheng Chen Efficient Efficient Retained

Liang Liu Efficient Efficient Retained

Stavros Harizopoulos Efficient Efficient Retained Qingbo Zhu Efficient Inefficient Retained

TABLE XI. APPLICATION AND REMARK

Project/Attributes Application level

Remark/ New invention Zichen Xu Very little Cost evaluation model and _{power evaluation model} Cheng Chen Very little

ReinDB – brown and green energy usage with additional

equipment to be installed Liang Liu Increasing

VM, Heuristic search algorithm and performance

sensitive

Stavros Harizopoulos Low Algorithm based and redesign the system component Qingbo Zhu et al. Low

Works only for multiple disk and focuses on high-end

storage system. IX. PROPOSED SOLUTION

Modules extracted from the analyzed research have been listed below and the integration is suggested in Figure 2.

Figure 2. Critical Solution Proposed on Energy Efficency and Green Computing Flow Model

1) Module 1: Supply driven Execution (SDE)

The core ideology is to align power consumption workload with the green supply through intertwined hardware component which requires advanced modeling on their performance and characteristic.

The scheduling is performed in the shortest slack time first manner. Within the slot, queries scheduled for excellent or same speed. The hardware is fixed within a slot whereby the energy consumption depends on the relative magnitude of the power consumption of the processing query.

2) Module 2: Integration of Solar-powered Database System

The integrated Solar-powered Database helps simulator or system take the solar energy trace and workloads trace as input, and replays the traces according to different execution strategies or methodology either battery, solar or carbon energy. It dynamically reschedules the workload and adjusts the hardware setting accordingly with the supply of throughput, and slack time preferred. Figure 3 explains how the execution of continuous energy supply takes place.

(6)

Figure 3. Solar Power Flow Model

3) Module 3: Disk Power Model & Management

Managing the multi disks speed to increase the amount of energy saved with data center workloads by identifying the active, idle and standby mode of the disk.

During the active mode, read and write mode is actively processing in the platters spin and the head. Likewise, during the idle, no disk is required whereby energy can be saved by idling the rotation and spinning. Mathematically, a lower rotational speed mode consumes lesser energy compare to higher rotations.

Reduction in Input/ Output Operations per Second (I/Ops) directly leads to reduction in the energy consumption which is shown in (1). An average latency is the time taken by the disk that is accessed to rotate to the position under read/ write head. Meanwhile the average seek time would be the time (ms) taken for the hard drive's read/write head to locate itself over the read/ write track.

I/Ops Calculation:

 seek time / 1000) + (latency / 1000) = I/Ops  

Energy Consumption= Power (Watt) x I/Ops

4) Module 4: Power-Aware Algorithms a) Power-aware offline Algorithm

This algorithm is an energy-optimal algorithm which replaces the block with the longest future reference distance.

b) Offline Power-aware Greedy Algorithm

This algorithm functions using the heuristic algorithm that consumes less energy.

5) Module 5: Power-aware Database Management System

It monitors the query optimizer, buffer manager, storage manager and many other database query optimization modules. Cost can be reduced by improving the query optimization time whereby the processing time can be improved and obtained through the setup of the server which is shown in (2). Cost estimation, C:

 C = (Query Optimization x time) x Processing Time 

6) Module 6: Green Cloud using Cloud Computing

It is to automatically make the scheduling decision on dynamically migrating or consolidating Virtual Machines among physical servers to meet the workload requirements while saving energy, especially for performance sensitive.

The heuristic algorithm is used in virtual machine migration to achieve the optimization. States of server power is adjusted through the dynamic voltage or frequency scaling.

X. FUTURE WORK

Traditional based database systems are fully functional to conciliate the mid-range of global market in database. The feature-driven approaches make the whole system complex indirectly increasing the energy wastage. Conversely, the future database system which is the Green Database should be able to anticipate and adopt the complexity of forthcoming algorithms to the multi-platform technologies architecture developed to address the market such as heterogeneous multicores, types of solid state technologies like USB flash drive, phase-change RAM, Electronic disk and other technologies which are compatible to multi-speed drives.

Full implementation and testing of the integration will eventually show the efficiency of the assimilation. Performance needs to be improved and results need to be stabilized. Database system needs to be made with more green-awareness. Not to disregard the data centers this is the major energy wastage source. We need to entail the abundant natural source of energy into technology to reduce the carbon energy usage. The awareness needs to be propagated and communities need to adapt and devise on this challenge focusing the transformation from performance-oriented research to energy-efficient computing through green computing.

XI. CONCLUSION

The green database implementation offers great potential in energy efficiency and green computing. This will not only reduce the energy wastage but also make existing running of data centers economically fit by reducing the cost and expenses. In this paper, problem of improving energy efficiency in data center as well as database management has been addressed from an integrated and new angle. Ultimately the aim and objectives are only one that is to minimize the wastage of energy and to retain the efficiency. Carbon footprint on database management needs urgent observation. We foresee that implementation of technology collaborated with the green energy will ultimately enhance the technology to be ecofriendly.

There is a lot of scope for exploring the future prospects in databases to make them greener in terms of energy saving. Our integration has some limitations but detailed research and involvement of the database community in this field can help improvise and lead towards a more efficient Green Database System.

REFERENCES

[1] L. A. Barroso, “The price of performance.” In ACM Queue 3, vol. 7, 2005, pp. 48-53.

[2] US Environment Protection Agency, “Report to Congress on Server and Data Center Efficiency: Public Law 109-431”, Energy star, retrieved June, 21, 2013, from http://www.energystar.gov/ia/ partners/prod_development/downloads/EPA_Datacenter_Report_Congr ess_Final1.pdf.

(7)

[3] US Department of Energy, Pacific Northwest National Laboratory, retrieved June, 28, 2013, from http://eioc.pnnl.gov/ research/sitawareness.stm.

[4] E. Mohammadi, M. Karimi, S. R. Heikalabad, (2011). “A novel virtual machine placement in cloud computing”, Australian Journal of Basic and Applied Sciences, 5(10): 1549-1555.

[5] Q. Zhu et al., “Reducing energy consumption of disk storage using power-aware cache management”, Department of Computer Science, University of Illinois at Urbana Champaign, 2004.

[6] S. Harizopoulos et al., “Energy efficiency: the new holy grail of data management systems research”, HP Labs & UCLA, 2009.

[7] Z. Xu, “Building a power-aware database management system”, University of South Florida, 2011.

[8] C. Chen et al., “Green databases through integration of renewable energy”, Nanyang Technological University & Sun Yat-sen University, 2012.

[9] L. Liu et al., “GreenCloud: a new architecture for green data center”, IBM China Research Laboratory: McGrill University, University of New Mexico, 2009.

[10] Xen User Manual, Citrix: Zen Server, retrieved June, 3, 2013, from

http://bits.xensource.com/Xen/docs/user.pdf.

[11] P. Ranganathan et al., “Models and metrics for energy-eﬃcient computing”, Advances in Computers, 75:159–233, 2009.

[12] K. Bansal et al., “A critical review on concept of green database”, Global Journal of Business Management and Information Technology, vol 1, pp. 113-118, 2011.

[13] D. Colarelli and D. Grunwald, “Massive arrays of idle disks for storage archives”, Nov 2002.

[14] S. Gurumurthi et al., “Interplay of energy and performance for disk arrays running transaction processing workloads”, In ISPASS, pp. 123– 132, Mar. 2003.

[15] S. Gurumurthi et al., “DRPM: Dynamic speed control for power management in server class disks”, ISCA, pp. 169–179, June 2003. [16] E. V. Carrera et al., “Conserving disk energy in network servers”, ICS,

June 2003.

[17] Amazon Elastic Compute Cloud (EC2), retrieved July, 2008, from

http://www.amazon.com/cc2/

[18] I. n. Goiri et al., “Greenslot: scheduling energy consumption in green datacenters”, International Conference for High Performance Computing, Networking, Storage and Analysis, SC ‟11, pp. 20:1–20:11, New York, USA, 2011.

[19] I. n. Goiri et al., “Greenhadoop: leveraging green energy in data-processing frameworks”, 7th ACM european conference on Computer Systems, EuroSys ‟12, pp. 57–70, New York, USA, 2012.

[20] A. Krioukov et al., “Integrating renewable energy using data analytics systems: Challenges and opportunities”, Bulletin of the IEEE Computer Society Technical Committee, pp. 1–9, 2011.

[21] Z. Liu et al, “Greening geographical load balancing”, ACM SIGMETRICS joint international conference on Measurement and modeling of computer systems, SIGMETRICS ‟11, pp. 233–244, New York, USA, 2011.

[22] A. Greenberg et al, “The cost of a cloud: research problems in data center networks”, ACM SIGCOMM Computer Communication Review, 39(1), pp. 68-73, 2008.