Parallel computing has become popular now-a-days due to its computing efficiency and cost effectiveness. However, in parallel computing systems, the computing demands a set of machines instead of a single machine. Therefore, it consumes a significant amount of power compared to single-machine computing systems. Moreover, a noticeable amount of power is necessary for maintaining the optimum temperature in the working environment of the parallel systems. This power is generally known as the cooling power required for the systems.
Although several power saving parallel computing schemes have already been proposed in the literature to date in order to minimize computational power consumption of a parallel system, designing a scheme considering both computational and cooling power consumption with low-cost resource is yet to be investigated in the literature. Therefore, in this thesis, we propose a low-cost power saving scheme simultaneously considering both computational and cooling power consumption. We design a machine learning framework BPGC, which tries to find the number of machines needed to be activated to be optimal, or at least near-optimal, in terms of minimum total energy consumption, with minimal overhead.
In order to predict total energy, we need to predict response time, computational power, and cooling power. We fit different machine learning algorithms for these predictions by using a year long collected training data. K-nearest neighbors, Support Vector Machine for regression, and Additive Regression using Random Forest show the highest accuracy for these predictions respectively. We implement BPGC framework in our test-bed with two green methods and static method. Our framework outperforms the green methods with a little degradation of QoS compared to the best QoS provider, that is, static method.
Difference Between Search & Browse Methods in Odoo 17
BGPC: Energy-Efficient Parallel Computing Considering Both Computational and Cooling Power Consumption
1. BGPC: Energy-Efficient Parallel Computing Considering
Both Computational and Cooling Power Consumption
Presented by
Tarik Reza Toha
#1205082
Department of Computer Science and Engineering,
Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
Defense Examination on B.Sc. Engg. Thesis
Supervised by
A. B. M. Alim Al Islam
Associate Professor
2. Overview of This Presentation
• Background and motivation
• Related work
• Proposed methodology
– Considering both computational and cooling power
consumption
• Performance evaluation
–Test-bed implementation
• Conclusion
2
3. Parallel Computing
3
Parallel computing refers to the use of multiple computational
machines (cores and/or computers) in combination to solve a single
problem
4. Applications of Parallel Computing
4
Many problems are so large and/or complex that it is impractical or
impossible to solve them on a single computer, especially given
limited computational memory.
Google now processes over 40,000 search queries every
second on average worldwide
Source: http://www.internetlivestats.com/google-search-statistics/
5. Applications of Parallel Computing [contd.]
5
Galaxy formation Planetary movements Climate changes
Modeling, simulation, and experimentation of complex real-world
phenomena demand rigorous computing
Traffic simulation Plate tectonics Weather forecasts
6. Energy Consumption of Data Centers
6
Data centers provide a significant number of computing systems
for parallel computing
Source: http://www.quotecolo.com/how-to-choose-the-best-green-cloud-hosting-provider/
7. Energy Consumption of DCs [contd.]
7
Cooling power is required for maintaining the optimum
temperature in the working environment of the data center
Source: Dayarathna, Miyuru, et al., 2016
8. Existing Power Saving Approaches
8
• Dynamic Energy-Aware Capacity Provisioning for
Cloud Computing Environments
– Zhang, Qi, et al., ICAC, 2012
– A homogenous cloud solution
• Provides optimum number of machines
– Trade-offs between energy efficiency considering
computational power and waiting time as QoS
• Considers the cost of turning on and off servers and fluctuation
in energy prices
Cooling power is
not considered!
9. Existing Power Saving Approaches [contd.]
9
• Power Management in Heterogeneous MapReduce
Cluster
– Sunuwar, Rojee, et al., 2016
– A heterogenous MapReduce cluster solution
• Addresses data unavailability of MapReduce cluster due to
DCP
– Trade-offs between energy efficiency considering
computational power and throughput as QoS
• Restrict CPU utilization of slave nodes
Cooling power is
not considered!
10. Existing Power Saving Approaches [contd.]
10
• Thermal Aware Server Provisioning And Workload
Distribution For Internet Data Centers
– Abbasi, Zahra, et al. HPDC, 2010
– A homogenous solution for Internet data center
– Trade-offs between energy efficiency considering both
computational and cooling power and response time as QoS
• Uses thermodynamic model of the data center
• Selects active servers based on least recirculated heat
Does not consider environmental weather impacts!
Inapplicable for co-located cluster environment!
Only applicable for spatially distributed cluster environment!
11. Our Contributions
We propose a machine learning framework,
BGPC, which predicts the number of machines for
minimum total energy consumption including
cooling energy while considering weather
conditions with minimal overhead
11
20. Failure Scenario of BGPC
20
̶ Equation of the total energy will be,
̶ The minimum point will be,
21. Conclusion
• The demand of data centers is increasing every year and, as
a result, the power consumption of data centers is also
increasing
– An energy-efficient parallel computing can be a good solution to
cope with this demand
• We provide a power saving scheme simultaneously
considering both computational power and cooling power
consumption with minimal overhead
– Outperforms existing greening method while maintaining QoS similar to static
method
• Future work
– Simulate BPGC in a large heterogeneous cluster
– Use context dependent classification in cooling power prediction (e.g. Kalman
filter)
– Use many objective optimization technique to optimize all QoS terms
21