This group research focus is resource sharing, control and management in wireless, fixed and computer networks. The members span virtual and physical infrastructures and networks and address mapping challenges between the virtual and physical worlds. The research addresses both virtual and physical networks mapping and global mobility challenges. The team considers that all objects can move and change location and this includes services and virtual machines. Infrastructures, hosting platforms and software and systems and environment dynamics are also taken into account in the investigations.
The group historically addressed personal, overlay, P2P wireless and fixed networks in addition to more traditional cellular networks operated by mobile networks and services providers. Our past work on personal virtual networks, distributed across private and personal clusters, led us naturally to address virtualisation of resources much like operating systems virtualisation is treated in computer systems and networks.
More recently, we have, unsurprisingly, combined our network background with the cloud computing paradigm to address joint service and network strata design in converged networks and clouds environments. The team aims first at removing the gap that exists between clouds and networks by addressing interfaces between these two realms and by designing the missing inter cloud networking services today. The second objective is to ensure that cloud principles and features such as on demand provisioning, elasticity and virtualisation are generalized to cloud and network federation instead of remaining confined to data centers. In essence, our main goal is to cover the computing, storage and communications dimensions jointly. Besides specifying new interfaces for clouds and networks interactions and coordination and designing software defined networking solutions and protocols, we derive novel virtual resources placement algorithms in physical infrastructures and hosting platforms using operations research methodologies, modeling and tools.
The group research spans theory and practice thanks to a “Networks and Clouds Federation” experimental platform that enables the validation of our models through deployment of our algorithms and solutions on the platform for performance evaluation, testing and consolidation. The platform provides advanced SDN technologies and the latest state of the art in cloud computing and cloud technologies. The modeling and optimization aspects focused on a number of key questions and issues in networks and clouds. The most salient ones address:
- Dynamic pricing in cloud federations,
- optimal placement of data and metadata in private and public storage,
- Energy -efficient exact allocation and consolidation algorithms,
- networking of distributed cloud resources,
- cloud networking interfaces and communicating middleware for distributed control and federation
3. Implementation and experimental assessments
The analytical work, analyses and designs are validated through experimentation using a dedicated clouds and networks emulation platform (depicted below) that enables realistic assessments, at scale, of the proposed resource allocation and management frameworks.
4. Details on the investigated topics
- Dynamic pricing in cloud federations involving multiple cloud providers each willing to cooperate by in-sourcing and out-sourcing demands to maximize their respective revenues and make efficient use of their resources. In this work we show that there is a bound (an optimal number in fact) in the number of providers that should be involved in a federation if they hope to obtain high revenues;
- Optimal placement of data and metadata in private and public storage using multi-commodity flow modeling to take into account frequency of reads and writes (PUT, GET, UPDATE, REMOVE) actions on stored data in order to dynamically optimize placement. The objective here is to optimally place data (especially placement groups) in storage spaces and create the duplicates and replicas according to experienced loads on the network and the storage system;
- Exact allocation and migration algorithms for energy-efficient VM Scheduling for Cloud Data Centers. The optimal allocation algorithm is solved as a bin packing problem with a minimum power consumption objective. The exact migration algorithm results from a linear and integer formulation of VM migration to adapt placement when resources are released. The objective is as expected to minimize energy consumption in data centers;
- Networking of distributed cloud resources from different cloud providers and sites is another topic the team is addressing from different angles. One of the key outcomes or achievements is the design and implementation of a Cloud Networking Gateway (CNG) that networks dynamically resources acquired from infrastructure providers using heterogeneous technologies and cloud managers. The CNG has been integrated with a cloud broker involving multiple users and providers in order to test and validate the gateway implementation in a realistic framework. The broker acts as an intermediary between tenants and providers and fulfills intermediation, arbitrage and aggregation functions. The CNG establishes the connectivity between the resources selected by the broker to compose the aggregates. This work is a component of a much broader study on cloud networking where the team develops exact and heuristic algorithms for Virtual Network Mapping and Cloud Networking from multiple providers. Inter provider connectivity is embedded when modeling the cloud networking problem. Simple forwarding nodes, programmable routers, servers and links (characterized or weighted by their latencies) are considered by the model. Exact and heuristic algorithms relying on network topology pattern have enable a factor or 100 to 1000 improvements in finding solutions to the problem.
5. Recent selected publications
I. Makhlouf Hadji, Djamal Zeghlache: Minimum Cost Maximum Flow Algorithm for Dynamic Resource Allocation in Clouds. IEEE CLOUD 2012: 876-882, 2012
II. Panagiotis Papadimitriou, Ines Houidi, Wajdi Louati, Djamal Zeghlache, Christoph Werle, Roland Bless, Laurent Mathy: Towards Large-Scale Network Virtualization. WWIC 2012: 13-25, 2012 (best paper award)
III. Mohamed-Haykel Zayani, Vincent Gauthier, Ines Slama, Djamal Zeghlache: Tracking Topology Dynamicity for Link Prediction in Intermittently Connected Wireless Networks. CoRR abs/1205.3328, 2012
IV. Ines Houidi, Wajdi Louati, Walid Ben-Ameur, Djamal Zeghlache: Virtual network provisioning across multiple substrate networks. Computer Networks 55(4): 1011-1023, 2011
V. Ines Houidi, Marouen Mechtri, Wajdi Louati, Djamal Zeghlache: Cloud Service Delivery across Multiple Cloud Platforms. IEEE SCC 2011: 741-742
VI. Houssem Medhioub, Ines Houidi, Wajdi Louati, Djamal Zeghlache: Design, Implementation and Evaluation of Virtual Resource Description and Clustering Framework. AINA 2011: 83-89
VII. Makhlouf Hadji, Wajdi Louati, Djamal Zeghlache: Constrained Pricing for Cloud Resource Allocation. NCA 2011: 359-365
VIII. Christoph Werle, Panagiotis Papadimitriou, Ines Houidi, Wajdi Louati, Djamal Zeghlache, Roland Bless, Laurent Mathy: Building virtual networks across multiple domains. SIGCOMM 2011: 412-413
6. Open source software produced by the group
- One “open source software” that has gained attention from the cloud community is an extension of the popular OCCI model produced by our group. PyOCNI (Python implementation of an Open Cloud Networking Interface) introduces cloud networking links and relationships originally missing in OCCI. The latter focuses only on compute and storage resources descriptions and links but neglects the communications dimension and the advanced networking specification essential to combine clouds and networks in a cohesive and comprehensive architecture. PyOCNI is an OCCI Server with HTTP and JSON rendering that relies on OCNI, the networking extension of OCCI. The source code and documentation are available at: https://github.com/jordan-developer/pyOCNI and https://github.com/jordan-developer/pyOCNI/blob/master/README.rst. Note that OCNI enables only interactions with networking services, the actual establishment or instantiation of nodes and links requires a dedicated gateway or network appliance. This is another facet addressed by the group that has also developed such a cloud networking gateway;
- The designed Cloud Networking Gateway (CNG) is an appliance acting as a connectivity establishment node (in our case software package in a VM) that can support cloud networking and clouds to interconnect virtual resources or machines co-located in the same data center or distributed across remote cloud sites be them public, private or hybrid. CNG can interact with cloud resource managers and networking services. It also handles naming and addressing requirements to establish connectivity on behalf of client applications or other components that need to establish connectivity dynamically and on demand.
- The group also developed another open source component to facilitate the exchange of information between multiple domains and especially networking domains. This is a Cloud Message Brokering Service (called CMBS) for exchanging messages in any N-to-N connection configuration or topology ( in any of the following fashion or type : req/rep, pub/sub, pipeline...). This component can be used for exchanging key information at several levels. It can be especially useful in :
- exchanging monitoring data,
- discovering or announcing services to feed the Service Provider Registry
- subscribing to any relevant event or service notifications
This component is also viable to support distributed control planes in establishing connectivity within a federation. In this case the objective is to exchange topology, addressing, naming and identity information at the networking level. A python implementation of CMBS, pyCMBS, developed under the Apache license, is available at: https://github.com/jordan-developer/pyCMBS. The development work continues with new releases regularly uploaded on the “github” as the code evolves.
7. Group members
Group leader: Prof. Djamal Zeghlache.
Members: Hadji Makhlouf, Houssem Medhioub are research fellows;
Doctoral students: Marouen Mechtri, Chaima Ghribi, Salma Rebai, Ines Fakhfakh, Houda Jmila; Research Assistants and Post Docs: Kaouther Drira, Francis Andreas Abrante, Sivasothy Shanmunaligam, Myriem Zekri
Posted 9 January 2013