GRENOBLE2011 - 2014I'm software developper in VMware Mobile Virtualization Platform team. As part of VMware Horizon Workspace project, we're providing an Android workspace for the Bring Your Own Device (BYOD) use-case. The workspace is based on an ARM hypervisor on which is running an Android virtual machine. Virtualization technologies ensure isolation between corporate and personnal applications and data.
As a developper, I'm working on guest Android paravirtualization (devices, OpenGL, etc.) and integration into host operating system.
Thales Avionics (Valence)
- Intern - Software development
Courbevoie2011 - 2011Integrated in the Global Navigation Satellite Systems (GNSS) department, I created then implemented a lightweight and robust file system for embedded flash (NOR) memories.
GNSS ground equipments were using NOR flash memories for firwmare, small data and crash reports but without any advanced management.
The aim of this file system was to:
- provide fault tolerance: be able to recover from power failures or system crashes without corrupting data (kind of transactions),
- enhance flash longevity using wear levelling.
Written in C, to be available to a large number of platforms (compilers were not available for all languages on other platforms), the project had to be compliant with DO-178B. Integrated to the team, I followed Agile principles (Scrum, XP and Lean) and used Test Driven Development.
Bull
- Intern - Software development
Les Clayes-sous-Bois 2010 - 2010I worked inside the "High Performance Computing benchmark team" to develop a monitoring tool to help understanding benchmarks behaviors on compute nodes.
The aim of the team is to provide the best answers to super computer calls for tender. The client provide a benchmark plus a goal, and the team has to propose the best super computer, for the lower price that fit the client request, in competition with other companies (IBM, NEC, etc).
My project consisted in develop a monitoring tool, based on Intel's CPU performance monitoring counters, to understand the behavior of a given benchmark measuring its use of the hardware (kind of instructions/operations, cache usage, memory bandwith, etc). But it had to as less intrusive as possible. Using the results, benchmarks can be optimized and their behavior extrapolated on bigger machines.
The tool, written in C, was based on a Linux kernel module to program and read the Performance Monitoring Counters of Nehalem architecture (some Model Specific Registers of Intel CPUs), and a user space reader to collect measures, export and exploit the results.
