- Designed and implemented a methodology to calculate fiber unit-cost (USD/strand/km/month) for Meta’s production network across the globe by solving a long-standing problem of calculating fiber length accurately (reduced error by 30%). This allows Network Planning team for the first time to produce optimal future network topologies based on real contracted unit-cost data which will enable savings of over $10M MRC.- Built pipelines for a centralized data warehouse of Meta’s global production network assets and fiber-optic cables (both terrestrial and subsea) that supports Network Infrastructure Key Performance Indicators (KPIs) and dashboards. This warehouse provides executives with critical insights to make informed investment decisions that support the scaling of Meta's ever-growing production network infrastructure. These insights include detailed information about Meta's network infrastructure, including capacity utilization, network performance, and cost per unit, helping executives to make data-driven decisions that optimize investments and ensure long-term success.- Developed a cost allocation methodology to attribute Edge CAPEX expenditures across a global network of distributed Edge Points of Presence (PoPs) and caching appliances, enabling granular visibility into equipment costs by location. This helps driving informed investment decisions by allowing to estimate the ROI of future Edge builds.- Developed an algorithm that identifies optimal locations for placement of decentralized cloud edge POPs (Point of Presence) in the US, in order to support latency-sensitive services such as VR/AR. This system aims to minimize latency for users on WiFi and Cellular networks and thus improve end-users’ experience by taking into account end users’ network performance, locations of production network assets and Meta’s existing Content Delivery Network (CDN) caches.

- Analyzed 4G coverage maps from 3 MNOs (Mobile Network Operators) across 12 countries as part of the groundbreaking 2Africa project, the largest subsea fiber-optic cable system ever made (45,000 km long). Developed and implemented data pipelines and scripts to automatically compare the MNOs’ coverage maps with the coverage maps generated through geospatial analysis of Facebook mobile users’ location datasets. Organized drive tests to provide even more precise insights related to signal strength and throughput. Results of my analysis proved the coverage gap that was worth more than 20M USD of additional network investment that partners of the consortium owed to Meta.- Built an impact ranking algorithm that determines the areas globally which simultaneously have high population density, poor network QoS (in terms of download speed and latency) and relatively low income. Such a system makes it more efficient to identify areas in the Emerging Markets that are strategically important for network interventions (fiberization, RAN upgrades, etc.) in terms of potential impact of the project on the global scale.

- Domain name recommendation system - conception and implementation of a search engine that recommends available domain names related to the input query by using various Machine Learning and NLP techniques- Data Analytics platform - responsible for setting up a cluster of nodes based on Elastic stack architecture used to collect and process data from various (structured and unstructured) data sources such as SQL databases, logs, JSON, CSV, text files or various APIs in order to statistically analyze numerous economic parameters of the company (turnover, cash sales, growth of sales per product, country or type of clients (individuals or professionals) on a daily, monthly and yearly basis)- Predictions of turnover and EBITDA - developed a simulator in Python that predicts turnover and EBITDA at the end of the current and the following years (prediction gets updated every minute) based on the sales to date and the predicted sales (calculated by applying linear regression techniques on invoices that are segmented based on several features).- developed software that calculates the accounted sales revenue (or deferred income) for all prepaid bills across four BUs (Business Units) by calculating the monthly impact of each bill during its commitment period. It calculates the reported sum of the deferred revenue for each month that corresponds to the impact of millions of bills (with 27 different currencies) from previous months and years.

The focus of the PhD thesis was the implementation of high-speed packet processing of TRILL protocol’s data plane (used to accelerate communication between Virtual Machines in the data center while reducing latency) on a massively parallel manycore processor (Kalray MPPA card – 256 processor cores) using ODP API. The thesis also researched the experimental evaluation of traffic latency, packet rate and throughput. Moreover, this thesis provides a mathematical analysis of the impact of different network topologies on the control plane’s load along with the state of the art on all the software-based and hardware-based fast packet processing solutions. Estimation of the CDF (Cumulative Density Function) of the packet's RTTs (Round-Trip Times) for various packet sizes shows that the proposed solution reduces latency significantly.

Cloud, Virtualization and Docker courses

Telecommunication systems

Design of a 3G/4G simulator in MATLAB programming language that estimates the joint impact of mobility and cooperation of base stations on quality of service (QoS) and performances for heterogeneous radio access networks 3G/LTE/LTE-A. The simulator takes into account different network parameters like propagation of radio waves, the spatial distribution of base stations, environment (urban, rural) but also the mobility of users. An analytical approach for modeling homogeneous and heterogeneous wireless networks (base stations and small cells or relays) has already been developed and it allows calculating performances in terms of throughput and to evaluate QoS in terms of probability of coverage or probability of blockage. The goal of the internship was to improve the present approach by modeling the movement of user “relays” and/or base stations which are relaying a useful signal to a final destination in a realistic environment (cities, buildings, streets, roads, etc.). Those users can be mobile terminals that are moving along with a vehicle like a car, train or metro and the station can for e.x. be placed on such vehicles.

System integration of the Service Delivery Platform solution (ECE, MSDP, EMG).