Working Together: Constraint Programming and Cloud Computing

Lead Research Organisation: University of St Andrews
Department Name: Computer Science

Abstract

This proposal combines two active, important research streams in cloud computing and constraint programming, both of which will realise significant and sustained benefits from working in concert. Constraint programming is a proven technology for solving complex combinatorial problems. However, the inherent difficulty of these problems means that performance can be variable, often requiring tuning by an expert to obtain best results. One approach to obtaining more robust performance is to employ a portfolio of solvers with complementary strengths. The scalable resource offered by the cloud is perfectly suited to the deployment of such portfolios and presents the opportunity to employ large solver portfolios to tackle challenge problems of exceptional difficulty. Conversely, a major concern in cloud computing is how to deploy an application on the available infrastructure so as to maximise performance and minimise operating costs. Added complexity arises when dealing with Big Data scenarios where it is important to run computation as closely (in terms of network distance) as possible to the data, in order to minimise network latency and maximise the performance of an application. This is a difficult combinatorial problem with a large set of variables including: public cloud provider, cloud configuration, geographical region, pricing etc. to which constraint programming is ideally suited.

Our two primary research streams in ICT will interact and work together with a third in astronomy to deliver a solution to a major challenge application: scheduling telescope observations to measure the abundance of planets throughout the Milky Way. If successful, the benefit to astronomy is clear, but our two primary streams will also benefit greatly from a major evaluation of their ability to work together to solve a large, complex problem.

Planned Impact

This proposal combines two active, important research streams in cloud computing and constraint programming, both of which will realise significant and sustained benefits from working in concert. Constraint programming is a proven technology for solving complex combinatorial problems and cloud computing is emerging as the dominant paradigm for hosting and delivering services over the Internet, both of which have very wide applicability. As a result of working together our research has a wide set of potential beneficiaries.

Impact to the Economy:

According to a recent Bloomberg report the total Worldwide cloud computing market could potentially be worth $270 billion by 2020. This project will foster collaboration between the complementary areas of cloud computing and constraint programming. The ideas and software produced as a result of this innovative collaboration will help ensure that the UK is on the forefront of the highly lucrative cloud computing market. In addition, the increased robustness of constraint solving gained from parallel constraint solver portfolio deployment, and the ability to scale constraint solving to problems of exceptional difficulty will be very valuable - combinatorial problems arise in many areas throughout the economy.

Regarding application and exploitation, there are currently no comprehensive services (discussed further in the case for support) that suggest the optimal public cloud provider and cloud configuration in order to maximise an application's performance and reduce operating costs. The technology developed as a result of our Working Together in ICT grant has the potential to be spun out into a cloud computing startup company and contribute directly to the UK economy. If this avenue is to be pursued we will source the correct government funding streams or venture capital investment.

Our Working Together in ICT project will likely be the start of a long lasting collaboration between the areas of cloud computing and constraint programming. Inward investment will be met with further follow on project sourced from UK (e.g., EPSRC), European (e.g., FP7) and International research funds (e.g., Qatar National Research Fund) as well as industrial sources.

Impact to People:

This project will employ two Research Assistants working at the intersection of cloud computing and artificial intelligence. The core team (P.I and Co-I's) will work closely with two Research Assistants funded directly from the project. This collaboration will generate new ideas around these diverse areas of Computer Science and help create a unique skill set which will lead to further research projects, training (through our workshops discussed in WP4.3) and potential knowledge exchange and furthering the training of skilled people for both academic and non academic disciplines.

Impact to Society:

Our major application focuses on gravitational micro lensing which provides the only technique known for detecting signatures of small planets (down to even Lunar mass) in wide orbits around their host stars. Embracing the amazing diversity of planets in the Milky Way is a requirement for understanding the origin, role, and possible fate of planet Earth. While comparative planetology beyond the Solar System addresses most fundamental philosophical questions, it also opens a window for gaining new insight on the interaction between ourselves and our planet, in particular on processes that shape the composition of the Earth's atmosphere, critical to life and the future of humanity.

Publications

10 25 50

publication icon
Akgun O. (2014) Breaking conditional symmetry in automated constraint modelling with CONJURE in Frontiers in Artificial Intelligence and Applications

publication icon
Akgun, O. (2014) Breaking Conditional Symmetry in Automated Constraint Modelling with Conjure in Proceedings of the 21st European Conference on Artificial Intelligence

publication icon
Akgun, O. (2013) Automated Modelling and Model Selection in Constraint Programming: Current achievements and Future directions in Workshop on Domain Specific Languages in Combinatorial Optimization

publication icon
Gent I (2017) Complexity of n-Queens Completion in Journal of Artificial Intelligence Research

publication icon
Gent I (2014) Generating custom propagators for arbitrary constraints in Artificial Intelligence

 
Description In a variety of settings we are faced with making decisions of increasing complexity and size, where large numbers of considerations interlock in complex ways. As an illustrative, consider the delivery of packages from a depot to a set of delivery locations by a fleet of vehicles. Packages must be assigned to vehicles while respecting vehicle capacity, driver shift patterns, and delivery deadlines, all while minimising the total distance travelled.

In this work, we have developed novel automated methods to model and solve problems of this type automatically and efficiently. Modelling is the process of expressing a problem to be solved in a format suitable to input to software that automates the search for solutions to decision-making problems. Automating this process is a particular challenge, and in this area we have advanced the state of the art.

The cloud marketplace offers a wide variety of on-demand resources with a wide range of performance capabilities. This makes it challenging for a user to make an informed choice as to which Virtual Machines (VMs) need to be selected in order to deploy an application to achieve maximum performance.

In this research project we have also developed methods to benchmark and compare VMs against one another in a single cloud provider. We also developed algorithms to schedule computational jobs expressed as Bag of Task (BoT) and workflows onto a cloud provider based on the metrics collected by the benchmarking.

Gravitational microlensing exploits a transient phenomenon where an observed star is brightened due to deflection of its light by the gravity of an intervening foreground star. In order to undertake efficient gravitational microlensing an observation schedule must be constructed such that various targets are observed while undergoing a microlensing event.

In this project we combined constraint programming, cloud computing and gravitational microlensing by developing a cloud-based e-Infrastructure that supports four methods to compute candidate schedules via the application of local search and probabilistic meta-heuristics. We then validated the feasibility of the e-Infrastructure by evaluating the methods on historic data. The experiments demonstrated that the use of on-demand cloud resources for the e-Infrastructure can allow better schedules to be found more rapidly.
Exploitation Route Complex and challenging decision-making and optimisation problems are central to a wide variety of important tasks. Therefore, any advances we make in the efficiency of their solution are valuable. This work provides important insights into how we can obtain substantial efficiency gains through improvements to the way in which we formulate models of the problems we wish to solve and provides a solid platform for future research in this area.

The cloud marketplace is becoming even more complex and competitive: there are more options from large players; new niche providers; and new instance types, e.g., container and cluster types. Our cloud benchmarking algorithms and methodologies are already being used by others to compare and contrast the vast array of public cloud provider's resource offerings. To cite and example: researchers have started to use our techniques to compare container instances, and deployments which container clusters (connected instances) of resources.
Sectors Aerospace, Defence and Marine,Agriculture, Food and Drink,Creative Economy,Digital/Communication/Information Technologies (including Software),Electronics,Energy,Financial Services, and Management Consultancy,Healthcare,Retail,Transport

 
Description The decision-making and optimisation software associated with this project is disseminated and supported via the website at constraintmodelling.org. It has found many and diverse applications around the world in industry, academia, and for pedagogical purposes. In finance we collaborated in the design and implementation of an automatic method of rebalancing an investment portfolio so that the relative weight of its constituent asset classes conforms to an idealised input model. We have collaborated with statisticians to develop a new statistical method to estimate the abundance of marine mammals (such as dolphins and porpoises) around offshore windfarms. This is helping to answer important questions about the impact of offshore windfarms on the number and distribution of marine mammals. Our software is used to improve the efficiency of the statistical method. Other illustrative examples include: modelling scheduling problems in a cloud-based project scheduling system; testing and debugging computer programs; level design in video games; and constructing and enumerating various structures in mathematics. The cloud benchmarking methods we developed throughout this project are general in scope, and can therefore can be applied to any cloud provider, e.g., Amazon or Google in order to aid resource selection based on performance. With the aid of an EPSRC Impact Acceleration Award (IAA) grant the benchmarking data from this research project was integrated into a web-based tool, which provides an ordered list (based on performance) of cloud providers virtual machine offerings when a user provides a set of high level application requirements. This has allowed our research work and the data it produced to be integrated into a tool which can be utilized by software engineers from around the world. Furthermore, the experiences from this grant helped Co-I Adam Barker obtain a Visiting Researcher position at Google in Mountain View, where he contributed to Google's internal cloud management platform.
Sector Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Financial Services, and Management Consultancy
Impact Types Economic

 
Description ABC: Adaptive Brokerage for the Cloud
Amount £386,558 (GBP)
Funding ID EP/R010528/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 04/2018 
End 03/2022
 
Description Impact Acceleration Account
Amount £3,576 (GBP)
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 03/2016 
End 03/2016
 
Description Impact Acceleration Account
Amount £38,309 (GBP)
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 07/2014 
End 12/2014
 
Description SFC Innovation Voucher
Amount £5,000 (GBP)
Organisation Government of Scotland 
Department Scottish Funding Council
Sector Public
Country United Kingdom
Start 10/2014 
End 03/2015
 
Description Standard Research
Amount £886,923 (GBP)
Funding ID EP/P015638/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 04/2017 
End 09/2020
 
Title Conjure 
Description An automated constraint modelling tool 
Type Of Technology Software 
Year Produced 2014 
Open Source License? Yes  
Impact This tool, still under active development, makes it possible to generate constraint models automatically from their specifications, rather than relying on human experts. 
URL http://constraintmodelling.org/conjure/
 
Title Minion 
Description A constraint solver 
Type Of Technology Software 
Year Produced 2006 
Open Source License? Yes  
Impact This software is in used by researchers worldwide to solve combinatorial search problems. 
URL http://constraintmodelling.org/minion/
 
Title Savile Row 
Description An automated constraint modelling tool 
Type Of Technology Software 
Year Produced 2014 
Open Source License? Yes  
Impact This tool is used for teaching and for research in various institutions worldwide 
URL http://savilerow.cs.st-andrews.ac.uk
 
Description All-Energy Glasgow 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Industry/Business
Results and Impact Our research group attended the All-energy exhibition in Glasgow in 2015, at which the University of St Andrews had a stand. We presented our work and engaged with a number of attendees at the show.
Year(s) Of Engagement Activity 2015
 
Description Demofest 2014 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? Yes
Geographic Reach Regional
Primary Audience Professional Practitioners
Results and Impact Presentation led to several discussions with potential industrial collaborators.

Demofest happened only very recently at the time of writing.
Year(s) Of Engagement Activity 2014
URL http://www.sicsa.ac.uk/knowledge-exchange/industry-collaboration/demofest/
 
Description IAA Showcase 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Industry/Business
Results and Impact This was a showcase event held at the University of St Andrews, presenting work from across the institution funded via EPSRC Impact Acceleration Accounts. It was open to the public and several colleagues from industry were invited.
Year(s) Of Engagement Activity 2015