Current Projects

If you like one of the projects listed below, please drop me a mail: lakis at elte dot hu

Real-time 3D point cloud processing in the edge [NEW!!!]

With the advent of Google Tango platform smart phones are not only able to capture images but using their in-built depth cameras 3D point clouds about their environment can also be captured. Currently, these point clouds are processed on the smart phones and used by various AR/VR applications. However, the computational complexity of these tasks requires very strong processors that has high energy consumption, leading to low battery life. By moving the processing to mobile edge computing nodes deployed close to the subscriuber, the processors of smart phones can be offloaded and thus the battery life can be extended. However, processing point clouds in an edge-cloud infrastructure may have further advantages (e.g. collaborative data collection and usage). On the other side, the strict delay requirement of real-time applications poses many challenges targeted by this project.

More information:

  • C++ knowledge - the project is based on open source components like PCL.
  • Implementation of existing algorihhms
  • Enhancements and modifications for cloud usage (based on Intel DPDK or other techniques)
  • Project launched this summer

Programmable abstract data planes for future networking solutions [Running]

In the past years a new paradigm called Software Defined Networking (SDN) has emerged that aims at simplifying many aspects of networking. The key idea behind SDN is the separation of data and control planes in switching devices and opening the control plane to be supervised by a logically centralized software controller, greatly simplifying the definition and deployment of innovative services and also the management of SDN devices. Current SDN solutions only focus on the programmability of the control plane, and consider the data plane as a collection of simple forwarding elements. However, this limitation of the data plane has many drawbacks, obstructing the development of optimal networking solutions and the rapid innovation in computer networks. In 2014, with the leadership of Jennifer Rexford (Princeton University) and Nick McKeown (Stanford University) a group of researchers proposed a new architecture and programming language called P4 for protocol independent packet processing. P4 is often regarded as a key milestone toward OpenFlow 2.0.

This research topic focuses on the new packet processing model of P4 and aims at answering different research questions from the following fields:

  • New network algorithms for traffic engineering solely in data plane e.g. traffic shaping
  • L4 load balancing in telecom and data centre networks
  • Utilization-aware routing
  • Data plane-driven TCP performance acceleration
  • Anomalous and malicious traffic behaviour detection and mitigation in data plane
  • Detection and diagnosis of performance problems by measurements in the data plane
  • Compilers for various platforms - NetFPGA (using Verilog, availabe at ELTE), Cavium (ARM-based networking device, GPU for packet processing, etc.)

More information:

Per Packet Value: Packet marking for advanced traffic engineering in telecom networks [Running]

Quality of Service is one of the most researched areas in networking literature. As usual for such a mature field, several solutions exist for most conceivable networking scenarios. However, the issues of QoS and resource management are still not fully solved problems - for example it is considered as a key issue in 5G standardization. Though many methods exist fulfilling certain needs, they are often focused on particular scenarios, limiting the possible application areas where they can be used.

In this project, we examine how a packet marking-based solution can be used as a QoS framework in future IP networks, what application scenarios can be solved easier than in existing frameworks, and how different - sometimes orthogonal - requirements can be satisfied in such networks. The students have to implement new concepts and algorithms for NS-3 simulator in C++.

More information:

H2020 MONROE - Reconstruction of operator policies in MBB networks for improved user experience: RECON

The complex Internet infrastructure plays a crucial role in determining end-to-end experiences of the users. The current Internet connects around 2 billion users through a global communication infrastructure that consists of thousands of service providers of different business types. A significant fraction of the users connect to the Internet through mobile broadband networks; using their notebooks, tablets and smart-phones. With the advent of 3G and 4G, a huge number of mobile applications have emerged with a wide range of requirements against the networking infrastructure. Ensuring high Quality of Experience has utmost importance for application developers. However, it is also well-known that network operators apply various policies in different layers of networking, affecting end-to-end traffic characteristics and eventually the observed Quality of Experience. Reconstructing and understanding these policies may help developers to make their applications compliant with the network and thus improve the provided QoE for end users. Such information can also be important for decision makers and supervisory bodies to get feedback about applied traffic management policies and practices, as well as the level of network neutrality to check their compliance with local regulations.

In this project, we aim at developing a methodology and toolboxes (as extensions) for MBB networks to reconstruct operator policies applied in Network, Transport and Application Layers, following a way of reverse engineering.

In network layer, we will mainly investigate intra and inter-domain routing policies with special focus on the stability of network paths and on the topology changes of mobile end-hosts. In transport layer, UDP and TCP functionality (e.g., UDT, MPTCP, ECN, TCP Fast Open) of different network paths will be tested through active measurements. The system allows experimenters to measure wide diversity of paths and can help determine whether a proposed solution has the required support or functionality from the MBB ecosystem. In application layer, operators often apply traffic differentiation—giving better (or worse) performance to certain classes of Internet traffic. This also provides vital information for the net-neutrality debate.

More information:

  • RECON is one of the Open Call 2 projects selected by MONROE for supporting
  • Check out MONROE site at