Distrinet, a new tool for distributed network emulation (2019-2020). Networks have become complex systems that combine various concepts, techniques and technologies. As a result, their modeling or simulation is now extremely complex and researchers make extensive use of prototyping techniques. Two experimental techniques are mainly used when testing a network: simulation and emulation. Emulation offers good accuracy and allows applications to be tested directly in an environment similar to a real environment. Most emulators do not take into account scaling, as they are usually designed to run in a single machine. Among other tools, Mininet is the most popular when it comes to evaluating SDN proposals. It allows SDNs to be emulated on a single machine. Unfortunately, Mininet shows its limitations for resource-intensive experiments because the emulation host can be overloaded. To solve this problem, we propose Distrinet, a distributed implementation of Mininet on multiple hosts. Distrinet uses the same API as Mininet, which means it is compatible with Mininet programs. It is generic and can deploy experiments in Linux clusters or in the Amazon EC2 cloud.
https://distrinet-emu.github.io/

Software tool and algorithms to optimze the execution time and scalability on requests on the database of the SAS MillionRoads (2019-2020). The tool was developped during a collaboration between Inria and MillionRoads for their product humanroads.
Code in Java acquired by the company.

Scheduling with network tasks (2018-2019). We consider the placement of applications (workflow or job) within a data center. Traditionally, this is done by a task orchestrator without taking into account network constraints. According to recent studies, network transfers account for up to 50% of the time required to complete traditional jobs. Thus, network resources must be taken into account when placing jobs in a data center. In [GHPT19], we propose a new scheduling framework, introducing network jobs that must be executed on network machines next to traditional CPU jobs. The model takes into account the communication competition for network resources, which is not taken into account in the previously proposed communication-based scheduling models. Network transfers within a data center can be easily modeled in our framework. As we show, classical algorithms do not efficiently manage a limited amount of network bandwidth. We therefore propose new algorithms with approximation guarantees in order to minimize the makepan in this framework. We demonstrate their efficiency and the importance of taking into account the network capacity through extended simulations on workflows built from Google data center traces.
https://gitlab.com/eesdn-others/scheduling-algorithm

Scalable routing and placement of service functions chains (2016-2018). Service chains consist of a set of network services, such as firewalls or application delivery controllers, that are interconnected by a network to support various applications. While this concept is not new, a very important new trend has emerged with the rise of software-defined networks (SDNs) and network function virtualization (NFV). The combination of SDN and NFV can make the service chain and application provisioning process much shorter and simpler. In [HJG18], we study service chain provisioning in conjunction with the number/location of virtual network functions (VNF). Although chains are often built to support multiple applications, the question arises as to how service chain provisioning can be planned to prevent data from passing through unnecessary network devices or servers and consuming additional bandwidth and CPU cycles. This requires careful selection of the number and location of VNFs. We propose an accurate mathematical model using decomposition methods with scalable solution to conduct such an investigation. We conduct extensive numerical experiments, and show that we can accurately solve the routing of service chain requests in minutes for networks with up to 50 nodes, and traffic requests between all node pairs. A detailed analysis is then performed on the best trade-off between minimizing bandwidth requirements and minimizing the number of VNFs and optimizing their location using different data sets.
https://gitlab.com/eesdn-others/sfc-cg

Interest clustering coefficients (2019-2020). We introduce a new metric to measure the clustering of a directed social graph with links of interest, namely the interest clustering coefficient. We calculated it (exactly and using sampling methods) on a very large social graph, a snapshot of Twitter with 505 million users and 23 billion links. We also provide the values of the previously introduced directed and undirected metrics, a first on such a large graph. We thus show that the interest clustering coefficients are higher than the classical directed clustering coefficients presented in the literature. This shows the relevance of metrics to capture the informational aspects of directed graphs. We also explored recommendations based on this new metric.
https://github.com/fredericgiroire/ICC

Design and optimized implementation of a probabilistic counting algorithm, MINCOUNT (2005-2006) (during my PhD, 2006). The algorithm in C has also been tested and validated by the research group of FRANCE TELECOM in Lannion, France, for its applications to network security.

Design and implementation of an algorithm MAPPING to build routes in a optical fiber network (patented algorithm) (2002) (internship in Sprint Labs, 2002). The software in C++ is used by the SPRINT Network Operation Center.