• de
Prof. Dr. Holger Karl

Research Topics

The research group Internet-Technology and Softwarization started in July 2021. We research Internet technology in general and for various specific forms of networks, e.g., mobile access networks, wireless networks for shorter ranges and more specific application areas, or data-centre networks. A common theme across these technological domains is so-called network softwarization. 

Topics of the research group

Network Softwarization

We are specifically interested in possible improvements by "softwarizing" a network: replacing built-in, hard-wired functionality by software, possibly even running on general-purpose hardware. This trend is partially driven by cost pressue and desire for flexibilty inside the network as such, which took form in concepts such as software-defined networking (SDN) or network function virtualization (NFV). These ideas were geared to network-internal functions like firewalls, load balancers, etc. 

But when actually thinking of the network as a programmable infrastrcuture and allowing general-purpose programs to be executed inside the network, new scenarios can be supported. A key advantage could be to reduce latency between users and services. To realize that, a software architecture like microservices is particularly attractive, dividing a complex service into smaller components that can be individually managed. A typical example would be a three-tier web application. 

This perspective allows us to think of many different kinds of applications and scenarios in a consistent manner. A typical example for a research question is then how to map a given application onto an available infrastructure. The applicatoin could be something very simple - like recording audio in a wireless network - or highly complex like video streaming with ad insertion; the infrastructure could be small and compact - just a few wireless nodes - or range up to an entire, nation-wide mobile access network. The basic ideas apply in all these cases; the details of the research questions certainly differ. 


Mobile networks

Mobile networks like 4G or 5G networks support mobile users, necessitating, e.g., handover and allocation of scarce resources (with radio spectrum, storage, data rate, ... all being relevant resources). Also, these networks operate at huge scales, with millions of users accessing hundreds of thousands of basestations in a country. Conventional optimization approaches have been well investigated, but keeping up with rapidly changing demands, user preferences, is challenging. Hence, looking at machine learning approaches for resource management is promising and considered a key ingredient for 6G networks, leading to self-driving networks. Also, conversely, using machine-learning-based applications inside mobile networks with their inherently volatile infrastructure is an increasing challenge.  

Wireless networks and cyber-physical systems

In this context, we consider the iodiosyncracies of wireless communication more in detail. A particularly interesting use case is the observation and control of a physical process - e.g., a machine on a factory shop floor - over wireless links. Here, stringent latency and dependability demands of control applications meet the time-varying, uncertain, difficult-to-predict nature of wireless channels, aggravated by the need to share these resources among multiple participants. Similarly to mobile networks, machine-learning approaches become more and more interesting. 

As a mid-term goal, we plan to establish a wireless control lab at HPI, providing experimental facilities to experiment with wireless technology and challenging control applications. Different use cases (e.g., including micro drons like Crazyflies) are currently under investigation. 


Data centres

Data centre networks typically carry different workloads than a general-purpose internet network. Typical scenarios include data-parallel applications like map/reduce. Resources of both the network and the servers need to be judiciously scheduled to obtain good performance (e.g., throughput, latency, energy efficiency).