On The Energy Efficiency Of Green Geo-distributed Datacenters

With rapid development and widespread applications of cloud computing over the past years,the core infrastructure — datacenters — which host diverse cloud applications,have been extensively deployed and operated all over the planet.For each of such geo-distributed datacenters,it typically consist of hundreds of thousands physical servers and consumes enormous energy,leading to tremendous electricity bill and carbon emissions.Under the global context of more completing cloud markets,severer energy starvation and deteriorative natural environment,the topic of how to efficiently manage the energy efficiency of geodistributed datacenters has quickly ascended to the spotlight of governments,industries as well as academia.To address the above challenge,existing literature mainly focus on reducing the energy consumption to improve the energy efficiency.Unfortunately,due to the spacial and temporal diversities of electricity carbon emission rate,reducing energy consumption does not necessarily means reduction of carbon emission,indicating that energy consumption and carbon emission should be co-optimized.To further green the cloud,renewable energy has been widely utilized in datacenters.However,for the emerging renewable energy of fuel cells,how to maximize their benefits in geo-distributed datacenters is still an open problem.On the other hand,datacenter demand response is envisioned as a promising approach to further reduce the energy cost.To materialize such benefits in geo-distributed datacenter,an efficient incentive mechanism for demand response is urgently required.Finally,in the era of big data,the frequent and large-scale data transmission among datacenters has made the inter-datacenter wide-ware network(WAN)the new bottleneck of energy efficiency,improving the energy efficiency of datacenter WAN is increasing critical for geo-distributed datacenters.To exploit the above new opportunities,a series of energy efficient techniques for geo-distributed datacenters are designed and analyzed,from the aspects of co-optimization of energy cost and carbon emission,optimized operation of emerging renewable energy,incentive mechanism for datacenter demand response,and energy-performance management for the cross-datacenter WAN.Carbon-aware power management for geo-distributed datacenters.How the spatial and temporal variabilities of the electricity carbon footprint can be fully exploited to further green geo-distributed datacenters is shown.Specifically,it is first verified that electricity cost minimization conflicts with carbon emission minimization,based on an empirical study of several representative geo-distributed cloud services.Then,a joint optimization on the electricity cost,service level agreement(SLA)requirement,and emission reduction budget is presented.To navigate such a three-way tradeoff in the presence of time-varying and unpredictable datacenter workload,a carbon-aware control framework is rigorously designed and analyzed by applying the Lyapunov optimization technique,which makes online decisions on geographical load balancing,capacity right-sizing,and server speed scaling.Optimized operation of fuel cell generation in geo-distributed datacenters.The demand for capping carbon emission has promoted the use of fuel cell energy in cloud computing,yet it is unclear what and how much benefit it may bring.To quantitatively examine the benefits brought by fuel cell generation,and to illustrate how such benefits can be realized with an intelligent coordination between grid power and fuel cell generation,a quantitative index UFC,called the utility of the cloud using fuel cells is proposed.UFC captures the level of the datacenters operator's overall satisfaction from energy cost,carbon emission,and workload performance.The UFC maximization problem is formulated to jointly optimize both fuel cell generation and geographical request routing.In order to avoid centralized solutions with high complexity and low scalability,a distributed algorithm blending the advantages of Alternating Direction Method of Multipliers(ADMM)is developed.An auction approach to pricing demand response of geo-distributed datacenters.Datacenter demand response is envisioned as a promising tool for mitigating operational stability issues faced by smart grids.It enables significant potentials in peak load reduction and facilitates the incorporation of distributed generation.Monetary refund from the smart grid can also alleviate the cloud's burden in escalating electricity cost.To incentivize the CSP participation,an auction mechanism which enables smart grids to voluntarily submit bids to the CSP to procure diverse amounts of demand response with different payments is presented.To maximize the social welfare of the auction,the CSP that acts as the auctioneer needs to solve the winner determination problem at large-scale.By applying the latest multiblock alternating direction method of multipliers,a distributed algorithm for each datacenter to solve a small-scale problem in a parallel fashion is proposed.Nice properties of the proposed auction,such as social welfare maximization and truthfulness are achieved by the Vickrey-Clarke-Groves(VCG)payment.WAN-aware Optimization for geo-distributed analytics.Geo-distributed analytics is a new computing paradigm in which analytics is performed across multiple geographically dispersed datacenters.While recent work on improving geo-distributed analytics has mostly focused on reducing either query response time or network traffic cost,it has been shown that there is a tradeoff between them.Navigating such a performance-cost tradeoff requires solving a joint optimization problem on the placement of input data and tasks,which is unfortunately NP-hard.Many previous solutions also assume full knowledge of future workloads,which is not practical in many cases.To address these challenges,Mensa— a scheduler that boosts geo-distributed analytics in a cost-efficient manner by scheduling data and task placement based on the bandwidth heterogeneity of the wide-area network —is presented.Mensa applies an offline heuristic to move data and an online greedy heuristic to place tasks.Mensa is the first scheduler that provides worst-case guarantees on both the query response time and WAN bandwidth usage,and without assuming perfect or full knowledge of future of the query.