Przemysław Szufel

On optimization of simulation execution on Amazon EC2 spot market

Abstract Large scale simulations require considerable amounts of computing power and often cloud services are utilized to perform them. In such settings the execution costs can be significantly decreased through the use of the Amazon spot price market. Its downside is that Amazon can interrupt the user’s computations when her bid price is too low. This poses a problem in finding an on-line bidding algorithm that balances the computation cost and the simulation experiment completion time. We identify key drivers governing the spot prices on Amazon EC2 and using these insights propose an adaptive bidding strategy that simultaneously minimizes the computation cost and the delays due to computation termination. We show that bidding close to a spot price and dynamically switching between instances is a strategy that is efficient and simple to implement in practice. In the paper we present a simulator of the EC2 spot pricing mechanism. The simulator can be easily used to develop and test other bidding strategies on Amazon spot price market.

A framework for sensitivity analysis of decision trees

In the paper, we consider sequential decision problems with uncertainty, represented as decision trees. Sensitivity analysis is always a crucial element of decision making and in decision trees it often focuses on probabilities. In the stochastic model considered, the user often has only limited information about the true values of probabilities. We develop a framework for performing sensitivity analysis of optimal strategies accounting for this distributional uncertainty. We design this robust optimization approach in an intuitive and not overly technical way, to make it simple to apply in daily managerial practice. The proposed framework allows for (1) analysis of the stability of the expected-value-maximizing strategy and (2) identification of strategies which are robust with respect to pessimistic/optimistic/mode-favoring perturbations of probabilities. We verify the properties of our approach in two cases: (a) probabilities in a tree are the primitives of the model and can be modified independently; (b) probabilities in a tree reflect some underlying, structural probabilities, and are interrelated. We provide a free software tool implementing the methods described.

D-Mason on the Cloud: An Experience with Amazon Web Services

D-Mason framework is a parallel version of the Mason library for writing and running Agent-based simulations – a class of models that, by simulating the behavior of multiple agents, aims to emulate and/or predict complex phenomena. D-Mason has been conceived to harness the amount of unused computing power available in common installations like educational laboratory. Then the focus moved to dedicated installation, such as massively parallel machines or supercomputing centers. In this paper, D-Mason takes another step forward and now it can be used on a cloud environment.

On parallel policies for ranking and selection problems

ABSTRACT In this paper we develop and test experimental methodologies for selection of the best alternative among a discrete number of available treatments. We consider a scenario where a researcher sequentially decides which treatments are assigned to experimental units. This problem is particularly challenging if a single measurement of the response to a treatment is time-consuming and there is a limited time for experimentation. This time can be decreased if it is possible to perform measurements in parallel. In this work we propose and discuss asynchronous extensions of two well-known Ranking & Selection policies, namely, Optimal Computing Budget Allocation (OCBA) and Knowledge Gradient (KG) policy. Our extensions (Asynchronous Optimal Computing Budget Allocation (AOCBA) and Asynchronous Knowledge Gradient (AKG), respectively) allow for parallel asynchronous allocation of measurements. Additionally, since the standard KG method is sequential (it can only allocate one experiment at a time) we propose a parallel synchronous extension of KG policy – Synchronous Knowledge Gradient (SKG). Computer simulations of our algorithms indicate that our parallel KG-based policies (AKG, SKG) outperform the standard OCBA method as well as AOCBA, if the number of evaluated alternatives is small or the computing/experimental budget is limited. For experimentations with large budgets and big sets of alternatives, both the OCBA and AOCBA policies are more efficient.

Distributed simulation optimization and parameter exploration framework for the cloud

Abstract Simulation models are becoming an increasingly popular tool for the analysis and optimization of complex real systems in different fields. Finding an optimal system design requires performing a large sweep over the parameter space in an organized way. Hence, the model optimization process is extremely demanding from a computational point of view, as it requires careful, time-consuming, complex orchestration of coordinated executions. In this paper, we present the design of SOF (Simulation Optimization and exploration Framework in the cloud), a framework which exploits the computing power of a cloud computational environment in order to carry out effective and efficient simulation optimization strategies. SOF offers several attractive features. Firstly, SOF requires “zero configuration”, as it does not require any additional software installed on the remote node; only standard Apache Hadoop and SSH access are sufficient. Secondly, SOF is transparent to the user, since the user is totally unaware that the system operates on a distributed environment. Finally, SOF is highly customizable and programmable, since it enables the running of different simulation optimization scenarios using diverse programming languages – provided that the hosting platform supports them – and different simulation toolkits, as developed by the modeler. The tool has been fully developed and is available on a public repository 1 under the terms of the open source Apache License. It has been tested and validated on several private platforms, such as a dedicated cluster of workstations, as well as on public platforms, including the Hortonworks Data Platform and Amazon Web Services Elastic MapReduce solution.

Asynchronous knowledge gradient policy for ranking and selection

The simulation of alternative evaluations in the ranking and selection problems often requires extensive amounts of computing power, so it is natural to use clusters with several workers for this task. We propose to extend the standard Knowledge Gradient policy to allow parallel and asynchronous dispatch of computation tasks among workers and denote it as the Asynchronous Knowledge Gradient. Simulation experiments indicate that performance loss due to parallelization of computations is below 25%. This implies that the proposed policy can yield significant benefits in terms of the time needed to obtain a desired approximation of the solution. We describe a master-slave architecture allowing for asynchronous dispatching of jobs among workers that handles problems with worker failures that are encountered in cluster environments. As a test bed of the procedure we developed an emulator of a heterogeneous computing cluster that allows testing of the parallel performance of stochastic optimization algorithms.

Statistical Verification of the Multiagent Model of Volatility Clustering on Financial Markets

Volatility clustering and leptocurtic, heavy tailed distribution of financial asset returns have been puzzling economists for decades. Ghoulmie, Cont, and Nadal (2005) proposed an agent-based model attempting to reproduce these stylized facts by means of the threshold switching behavior of investors. We investigate properties of the model following principles of the design of simulation experiments. We find the results to be only partially consistent with properties of empirical time series. This suggests the model to be an insightful but incomplete description of the phenomena under study.

Optimal execution of large scale simulations in the cloud: the case of ROUTE-TO-PA SIM online preference simulation

Cloud computing enables massive parallelization of execution of large scale simulation experiments but it is complex to do it in a cost-efficient way. We present methodology used to achieve this goal that was devised in the ROUTE-TO-PA project, where we develop a simulator for generalization of the dynamics of preferences observed on the social platform to the entire population. Experimenting with such a complex simulation model over a computing cluster in the cloud requires solving not only technical challenges (solution architecture and management of dynamically changing infrastructure) but also requires optimization of computing cost. In this work we present our approach (ROUTE-TO-PA SIM) to configure and manage such environment in the Amazon Web Services cloud setting.

On elicitation of preferences from social networks through synthetic population modelling

Social network platforms are a useful source of information on preferences of citizens. However, population exposed on social network platform is non representative and in result preferences collected through such platforms are biased. The goal of the public administration is to utilize the data that can be collected through such online platforms in order to understand preferences and its structure in the society and hence better react to communitys needs. This situation calls for an algorithm that will allow to generalize information collected on social platform users on the entire population. We propose and evaluate a two-step methodology for testing of such algorithms: (1) synthetic population is generated and its sample is selected that represents social platform users and (2) regenerate the whole population on the basis of data from sub-population. In this way we can evaluate the quality of different algorithms aimed at preference elicitation from social platform data.

ROUTE-TO-PA H2020 Project: Raising Open and User-Friendly Transparency-Enabling Technologies for Public Administrations

In this short paper, we introduce ROUTE-TO-PA project, funded by European Union under the Horizon 2020 program, whose aim is to improve the transparency of Public Administration, by allowing citizens to make better use of Open Data, through collaboration and personalization.