Research on Frontier Technologies in Data Center and Server
With the continuing growth of the digital economy and the steady promotion of New Infrastructure Strategy, internet data centers (IDCs) have become fundamental utilities to support China’s socio-economic development. The energy demand of IDC industry, in particular, is sustainably increasing, attracting more and more attention. In this case, energy efficiency and elastic scheduling of flexible workloads is not only prerequisites for the sustainable development of IDC industry, but also the way to assist in achieving the national commitment of “3060” carbon neutrality.
Technically, the execution of some workloads can provide scheduling flexibility potential. In specific, delay-intensive workloads can be postponed in time scale considering their lower priority. Delay-sensitive services can be spatially shifted to another server cluster for processing. Considering the electricity production and carbon emission changing with time and space, the potential of shifting workloads on temporal and spatial dimensions can be utilized for cost reduction, energy saving and carbon-free operation for IDC industry. For achieving this goal, it is essential to explore the capability of IDCs to change the power demand by elastic scheduling of workload and the sensitivity to follow the price and emission signal released from power systems. The coordinated operation of power systems and IDCs can contribute to real-time balance in energy supply and demand and reduce carbon emission. In particular, the current released strategy “East Data Computing in West”, representing moving east data to west areas, further emphasized the necessity of this work.
In Alibaba Group, Alimama team, which is in overall charge of the advertising business, has successful practical experience in shifting advertising streaming workloads spatially. Moreover, the advertising jobs include both delay-intensive and delay-sensitive workloads. Therefore, in this research, we intend to cooperate with Alimama team in exploiting the feasibility of spatial and temporal scheduling of advertising jobs theoretically and practically, such as scheduling algorithm design, power consumption model establishment, and practical implementation. This research project is the first step of “East Data Computing in West” implementation. And, it also makes Alibaba the first promoter of carbon reduction by workload scheduling in China. The pilot experience of this project will be replicable in China.
In addition, this project will cooperate with Jibei power grid, Jiangsu power grid, Zhejiang power grid, and western Inner Mongolia power grid in fully awareness of the cost and resource difference. This project can increase the renewable integration and quantify the ability to reduce IDC industries in different regions.
This project aims to propose a quantitative model for IDC workload scheduling that considers the actual operation requirements of workloads and scheduling systems. Then, the proposed model is applied in a power market scenario to measure the carbon reduction that IDCs can achieve by workload scheduling. Finally, the proposed model will be used in some pilots to participate in electric markets, electricity ancillary services markets, or demand response markets to evaluate the effectiveness of the proposed model in cost saving and carbon reduction. And the performance of IDC participating in the cooperation with power systems is also quantified.
This project proposes to conduct an in-depth research on three technical problems:
- A mathematical model of our current scheduling system, including resource allocation and virtual machine allocation of different kinds of workloads.
- A simulation model of the cooperation of IDCs and the power system, considering workload scheduling and power system operation
- A scalable testing environment for testing the reliability and feasibility of the cooperation model
Related Research Topics
- Analysis on the scheduling boundaries based on the current requirements of scheduling systems for different kinds of workloads.
- A data-driven model of the mapping relationship between workloads execution and power load of servers or IDCs.
- A cooperation model between IDCs and power systems considering workload elastic scheduling, considering actual scheduling boundaries.
- Detailed implementation plans of IDC participation in the power markets, electricity ancillary service markets, or demand response markets for our different IDC bases.
- Evaluate the performance of the proposed model in carbon and cost reduction for different IDC bases.
- Sensitivity analysis of electricity price, energy resources, and carbon intensity on the performance of the proposed model in carbon and cost reduction.