RIT | Rochester Institute of Technology

Networking and Security Research Group

Project 1: Machine Learning and Network Research on the NSF FABRIC Testbed

The RIT Network Research team is combining machine learning techniques and topologies built on the NSF funded FABRIC infrastructure in order to address networking and security challenges. These are important research areas but equally important is developing the related student skills and knowledge.

Current experiments require machine learning nodes containing models, datasets for training, and a workflow moving a trained model to an operating topology housed on the virtualized testbed. This operation and the components embody a collection of advanced concepts that can make it difficult for students to seamlessly join machine learning-based projects. While students understand many of the fundamental components (networking, virtualization, coding, machine learning), research work performed on a complex FABRIC infrastructure is much larger in size and scope.

This project seeks to engage a pair of students in our current machine learning, networking, and security work in order to determine what they need to learn and what entry-level skills they should possess. This information will be used to prepare the next generation of students working on this or related projects. It will also provide needed input to future funding proposals.

Specifically, the goals of this project include:

● Have students work from start to finish on a project housed in FABRIC in order to understand and document the student perspective. This will enable us to better understand important prerequisite knowledge and what learning experiences must be created.

● Outline a project plan and operational requirements for externally funded solicitations such as those found in the NSF NetS program which involves both I-School and NTID students.

Current research questions under examination include:

● Packet classification via neural network ensembles

● ARP poisoning and adversarial attacks

● Connectivity outages and topology management

● Dataset creation

NSF FABRIC (FABRIC is Adaptive ProgrammaBle Research Infrastructure for Computer Science and Science Applications) is an International infrastructure that enables cutting-edge experimentation and research at-scale in the areas of networking, cybersecurity, distributed computing, storage, virtual reality, 5G, machine learning, and science applications.

The FABRIC infrastructure is a distributed set of equipment at commercial collocation spaces, national labs and campuses. Each of the 29 FABRIC sites has large amounts of compute and storage, interconnected by high speed, dedicated optical links. It also connects to specialized testbeds (5G/IoT PAWR, NSF Clouds), the Internet and high-performance computing facilities to create a rich environment for a wide variety of experimental activities. FABRIC Across Borders (FAB) extends the network to 4 additional nodes in Asia and Europe.

An Experiment Topology and FABRIC Resources

Current Undergraduate Researchers

Gavin Hunsinger

· Fourth Year Computer Information Technology Student

· Strong interest in networking and cyber security

· Utilized FABRIC's virtual Testbed to collect data and experiment with FRRouting

· Created Python scripts to gather telemetry data and parse Wireshark packets

· Started developing a neural network to classify types of packets

· Python tools: pytorch, CUDA

Faria Sultana

· 2nd year student majoring in Computing and Information Technology, interested in networking and system administration. 

· Used the Fablib API to build layer 3 topologies, wrote Python scripts to conduct bandwidth testing, documented the process

· Collected packet datasets and preprocessed them to be injected into a neural network model

· Learning how a neural network works and how we can improve on it

New team members

Zachary Riback and Jonathan Bateman

Previous team members

Amira Chhaiouine

· Undergraduate rising third year cybersecurity student.

· Highly interested in cybersecurity, networking, and machine leaning

· Utilized the FABRIC virtual testbeds to create L2 and L3 topologies and performing different experiments

· Tools: iperf, nmap, FRRouting, ping, traceroute, Wireshark

· Creating python scripts to gather telemetry data in order to parse Wireshark packets.

· Began the development of a neural network to classify packet types.

Faculty Investigators: Bruce Hartpence, Daryl Johnson, Bill Stackpole

Contact: bhhics@rit.edu

Project sponsors: RIT Research Computing and the Golisano College of Information Sciences

Project 2: RIT Network and Security Dataset Repository

The lack of quality datasets is a critical problem that limits machine learning experimentation in the areas of intelligent networks and security. This collection of datasets contains packets from a variety of contemporary (2020 onward) layer 2, layer 3, layer 4 and application layer protocols. It is a growing series of curated and un-curated datasets for both general network classification experiments and security investigations. The curated datasets are used as part of our machine learning classification work and the collection contains some details on model structure and baseline accuracy for comparison.

This collection contains curated (ex. enforced balance) and non-curated (raw) datasets in several formats.

This works was recently published in the International Symposium on Networks, Computers and Communications (ISNCC'24)

Links:

Network datasets: https://repository.rit.edu/data/1/

When referencing the networking datasets, please use the following DOI:

10.57673/gccis-yg55 OR www.doi.org/10.57673/gccis-yg55

Security datasets: https://repository.rit.edu/data/2/

When referencing the security datasets, please use the following DOI:

10.57673/gccis-qj60 or https://doi.org/10.57673/gccis-qj60

Contact info:

Bruce Hartpence bhhics@rit.edu

Daryl Johnson dgjics@rit.edu

Bill Stackpole wrsics@rit.edu