Ric D. Herbert

Resource allocation to conserve energy in distributed computing

Energy consumption is an issue in grid computing. There has been substantial research into grid resource allocation, but little research on energy aware resource allocation. We propose that altering the resource allocation mechanism to incorporate node power and performance data can make a substantial difference to both the time taken to execute tasks and the energy consumed by the grid. This paper examines the use of three simple economic resource allocation mechanisms through simulation. We discover that different mechanisms perform better under different circumstances, and that changing the resource allocation mechanism to incorporate the power and performance information of individual nodes can result in a substantial difference to the time taken to execute tasks, and over time can make a marked difference to the total energy consumption of the grid resource.

Constrained macroeconomic policy development with a separate predictive model

This paper examines the issue of the generation of optimal control policies where there are explicit constraints upon the control values and there is limited knowledge of the complex economic system. The paper develops a methodology where the constrained optimal control is based upon a separate model that predicts the policy targets for the economic system. The methodology as applied to a small calibrated macroeconomic model of Australia.

Reducing energy consumption in distributed computing through economic resource allocation

Energy consumption is an increasingly important consideration in computing. High-performance computing environments consume substantial amounts of energy and the cost of energy is increasing. We explore the possibility of reducing the energy consumption of a grid of heterogeneous computers through appropriate resource allocation strategies. We examine a number of possible grid workload scenarios and analyse the impact of different resource allocation mechanisms on energy consumption and time taken to execute tasks. We perform this analysis first on a cluster of heterogeneous nodes and then scale up the experiment to a grid of multiple clusters. Our results show that different resource allocation mechanisms perform better under different scenarios, and that selection of the resource allocation mechanism can significantly alter grid energy consumption.

Reducing grid energy consumption through choice of resource allocation method

Energy consumption is an increasingly important consideration in computing. High-performance computing environments consume substantial amounts of energy, at an increasing financial and environmental cost. We explore the possibility of reducing the energy consumption of a grid of heterogeneous computers through appropriate resource allocation strategies. We examine a number of possible grid workload scenarios and analyse the impact of different resource allocation mechanisms on energy consumption. We perform this analysis first on a cluster of heterogeneous nodes, then on a grid of several clusters. Our results show that different resource allocation mechanisms perform better under different scenarios, and that selection of an appropriate resource allocation mechanism can significantly reduce the total grid energy consumption.

Predictive Constrained Policy Generation for Macroeconomic Systems

Abstract This paper examines the generation of optimal control policies where there are explicit constraints upon the control values and their rates of change, and there is limited knowledge of the complex economic system. The paper develops a methodology using quadratic programming where the constrained optimal control policies are based upon a learning model that predicts the policy targets for the economic system. A subset of the control policies is applied to the economic system and from the response of the system a new predictive model and resultant optimal controls are generated. The methodology is then repeated. A numeric example of the methodology as applied to a macroeconomic model is presented in the paper.

Clustering Obsolete Computers to Reduce E-Waste

Personal computers contribute significantly to the growing problem of electronic waste. Every computer, when finished with, must be stored, dumped, recycled, or somehow re-used. Most are dumped, at a huge cost to health and the environment, as their owners succumb to the desire to keep up with the ever-increasing power of new computers. Supercomputers and computer clusters provide more power than ordinary desktop and laptop computers, but they too are subject to rapid obsolescence. The authors have built a cluster of obsolete computers and have found that it easily outperforms a fairly standard new desktop computer. They explore how this approach can help to mitigate e-waste, and discuss the advantages and limitations of using such a system.

Assessing two common approaches for solving models with saddle-path instabilities

This paper presents an approach for assessing the time taken by the well known reverse-shooting and forward-shooting algorithms to solve large-scale macroeconomic models characterized by saddle-path instability. We focus on a range of investment models with multi-dimensional specifications of the capital stock. Each algorithm presents a complicated exercise with a potentially unstable ordinary differential equation to be solved over a wide parameter space and involving a difficult search. Our results provide insights into how the complexity of the solutions to a broad range of macroeconomic models increases with the dimensionality of the models. We describe how econometric techniques could be used to summarize the likely success of competing algorithms when confronted with models exhibiting a range of properties.

Exploiting Model Structure to Solve the Dynamics of a Macro Model

This paper considers alternative numerical approaches to solvingthe time-path of a nonlinear representative agent model.

Predator-prey systems depend on a prey refuge.

Models of near-exclusive predator-prey systems such as that of the Canadian lynx and snowshoe hare have included factors such as a second prey species, a Holling Type II predator response and climatic or seasonal effects to reproduce sub-sets of six signature patterns in the empirical data. We present an agent-based model which does not require the factors or constraints of previous models to reproduce all six patterns in persistent populations. Our parsimonious model represents a generalised predator and prey species with a small prey refuge. The lack of the constraints of previous models, considered to be important for those models, casts doubt on the current hypothesised mechanisms of exclusive predator-prey systems. The implication for management of the lynx, a protected species, is that maintenance of an heterogeneous environment offering natural refuge areas for the hare is the most important factor for the conservation of this species.

Calculating Short-Run Adjustments: Sensitivity to Non-Linearities in a Representative Agent Framework

Two common properties of macroeconomic models are non-linearities and dynamics characterised by a non-zero number of unstable eigenvalues. Under these circumstances, a common approach is to make analysis more tractable by linearising the model in the neighbourhood of an appropriate steady-state. The linearised model is then employed to calculate short-run adjustments following exogenous shocks. This can lead to different results than would be derived from the correct (non-linear) model. This paper investigates the magnitude of errors that come about as a consequence of using a linear approximation to a well-known representative agent model. This is achieved by taking a calibrated version of the Matsuyama (1987) model of a small open economy.