Characterizing big data analytics workloads on POWER8 SMT processors

Big data analytics is emerging as one kind of the most important workloads in modern data centers. Hence,it is of great interest to identify the method of achieving the best performance for big data analytics workloads running on state-of-the-art SMT( simultaneous multithreading) processors,which needs comprehensive understanding to workload characteristics. This paper chooses the Spark workloads as the representative big data analytics workloads and performs comprehensive measurements on the POWER8 platform,which supports a wide range of multithreading. The research finds that the thread assignment policy and cache contention have significant impacts on application performance. In order to identify the potential optimization method from the experiment results,this study performs micro-architecture level characterizations by means of hardware performance counters and gives implications accordingly.

Power Flow Analytical Solutions and Multi-dimensional Voltage Stability Boundaries Based on Multivariate Quotient-difference Method

This paper proposes a novel Multivariate Quotient-Difference(MQD)method to obtain the approximate analytical solution for AC power flow equations.Therefore,in the online environment,the power flow solutions covering different operating conditions can be directly obtained by plugging values into multiple symbolic variables,such that the power injections and consumptions of selected buses or areas can be independently adjusted.This method first derives a power flow solution through a Multivariate Power Series(MPS).Next,the MQD method is applied to transform the obtained MPS to a Multivariate Pad´e Approximants(MPA)to expand the Radius of Convergence(ROC),so that the accuracy of the derived analytical solution can be significantly increased.In addition,the hypersurface of the voltage stability boundary can be identified by an analytical formula obtained from the coefficients of MPA.This direct method for power flow solutions and voltage stability boundaries is fast for many online applications,since such analytical solutions can be derived offline and evaluated online by only plugging values into the symbolic variables according to the actual operating conditions.The proposed method is validated in detail on New England 39-bus and IEEE 118-bus systems with independent load variations in multi-regions.

Approximate Controllability of Neutral Functional Differential Systems with State-Dependent Delay

This paper deals with the approximate controllability of semilinear neutral functional differential systems with state-dependent delay. The fractional power theory and α-norm are used to discuss the problem so that the obtained results can apply to the systems involving derivatives of spatial variables. By methods of functional analysis and semigroup theory, sufficient conditions of approximate controllability are formulated and proved. Finally, an example is provided to illustrate the applications of the obtained results.

Prediction of voltage distribution using deep learning and identified key smart meter locations

The energy landscape for the Low-Voltage(LV)networks is undergoing rapid changes.These changes are driven by the increased penetration of distributed Low Carbon Technologies,both on the generation side(i.e.adoption of micro-renewables)and demand side(i.e.electric vehicle charging).The previously passive‘fit-and-forget’approach to LV network management is becoming increasing inefficient to ensure its effective operation.A more agile approach to operation and planning is needed,that includes pro-active prediction and mitigation of risks to local sub-networks(such as risk of voltage deviations out of legal limits).The mass rollout of smart meters(SMs)and advances in metering infrastructure holds the promise for smarter network management.However,many of the proposed methods require full observability,yet the expectation of being able to collect complete,error free data from every smart meter is unrealistic in operational reality.Furthermore,the smart meter(SM)roll-out has encountered significant issues,with the current voluntary nature of installation in the UK and in many other countries resulting in low-likelihood of full SM coverage for all LV networks.Even with a comprehensive SM roll-out privacy restrictions,constrain data availability from meters.To address these issues,this paper proposes the use of a Deep Learning Neural Network architecture to predict the voltage distribution with partial SM coverage on actual network operator LV circuits.The results show that SM measurements from key locations are sufficient for effective prediction of the voltage distribution,even without the use of the high granularity personal power demand data from individual customers.