Contact: Bernard Fortz bfortz@ulb.ac.be
Luce Brotcorne : Luce.Brotcorne@inria.fr
Inria Lille Nord Europe, INOCS Team
To apply send a CV and a cover letter to B Fortz and L. Brotcorne before April 26.
Context
Cloud computing is offering a variety of services to end users like the management and storage of data, the processing of jobs or the access to platforms on demand.
The service level agreement (SLA) is the contract between the service provider and the customers. It defines the services that the provider will furnish. The quality of service (QoS) represents the capacity of the service provider to respect the SLA subscribed by the clients. Focusing on the "Platform as a service" or the Infrastructure as a service" context, the QoS can be deteriorated if the delay of processing increases. This occurs when the amount of resources required by the users are not sufficient to satisfy the demand.
To solve this problem the cloud provider can either reduce the maximum resources consumption of users in the SLA or invest in additional servers or define incentives like the price to smooth out the demand over time. This last solution is under study in this project.
When defining the service prices, the cloud service provider objectives are to increase the benefits while insuring a good quality of service. Three pricing strategies can be identified in the literature : i) value-based pricing, based on the demand of users, ii) cost-based pricing, based on the costs for the service provider, iii) market-based pricing considering both aspect, like auction model for example.
Unfortunately, these "pay-per-use fixed pricing" charging users for what they consume can't be used to limit the peak periods of usage by the users. In order to reach this objective, prices need to vary over time and according to the amount of resources required.
To intrinsically integrate the decisions of the cloud users maximizing their utility into the decision making process of the cloud service provider the cloud service pricing problem (CSPP) can be modeled as a bilevel optimization problem.
Bilevel Programming is a fairly recent branch of optimization that deals with programs whose constraints embed an auxiliary optimization problem . More precisely bilevel problems involve two decision makers (a leader and a follower) interacting sequentially and hierarchically. In our context the leader is the cloud operator defining a pricing strategy taking explicitedly into account the reactions of the users. For the CSPP the objective of the leader is to maximize the revenue (profit -costs) and decrease the peaks while the objective of the users is to minimize their cost and they delay.
Bilevel programming problems, being generically difficult to solve due to their non-convexity and non differentiability, the structure of the problem will be exploited to define efficient solution methods. For example when the optimization problem is convex for fixed leader decisions, it can be replaced by its KKT conditions leading to a single level optimization problem.
- - Research objective--
The goal of the post-doc is to study the properties of bilevel bilinear programs for the SCPP and develop efficient solution methods. Numerical results should be studied and discussed. This field of research is and very innovative and promising in an industrial context.
Where
I NRIA Lille Nord Europe, INOCS Team
The INOCS team aims to develop new models, algorithmic techniques and implementations for optimization problems with complex structure (CS). More precisely, we consider that an optimization problem presents a CS when for example it involves some hierarchical leader-follower structure (bilevel optimization). Luce Brotcorne is specialist in bilevel optimization with a particular expertise to solve pricing problems, while Bernard Fortz has also a strong experience in decomposition methods that will be at the core of algorithms developed in the project
A post doctoral position is available in the Inocs team for the topic ``Services Pricing for cloud computing''.
Knowledge
Candidates should hold a PhD Thesis in Operations research, mathematics, computer science, or similar fields and should ideally have a solid background in discrete optimization, integer programming, decomposition techniques. Computer science skills in algorithmic and C/C++ development are also welcome.
Knowledge of French is not required, but good communication skills and a solid knowledge of English are essential.
**********************************************************
*
* Contributions to be spread via DMANET are submitted to
*
* DMANET@zpr.uni-koeln.de
*
* Replies to a message carried on DMANET should NOT be
* addressed to DMANET but to the original sender. The
* original sender, however, is invited to prepare an
* update of the replies received and to communicate it
* via DMANET.
*
* DISCRETE MATHEMATICS AND ALGORITHMS NETWORK (DMANET)
* http://www.zaik.uni-koeln.de/AFS/publications/dmanet/
*
**********************************************************
Posts
