Main Content

Flow caching for supporting efficient privacy-aware publish/subscribe over programmable switches

Publish/subscribe is a key paradigm for supporting scalable communication between producers (publishers) and consumers (subscribers) of information. To maintain decoupling between publishers and subscribers, a broker network helps in routing event messages. To support low latency in routing events, programmable network devices are suitable to perform routing decisions directly on the header messages of packets, reducing the communication path and avoiding the overhead in the operating system completely to deliver the packets to event brokers. While in-network publish/subscribe enforces high efficiency, it typically sacrifices privacy since every packet requires carrying routing information that, in parts, also discloses the content of the messages once an adversary understands the schema of attributes.

A way to improve privacy is to give each of the routers a personalized/individual view of the event schema, i.e., the order of attributes of which an event is comprised.  Here, each router receives packets subject to a random permutation of the order of attributes, thereby making it hard for an adversary to understand the content of the header data. Unfortunately, this method also has a drawback because it requires each broker to handle a significant number of rules to prevent adversaries from understanding the content of packets. However, in switches, the number of rules that can be processed fast is limited. 

The goal of the thesis is to design a concept for caching flow table entries to enhance the efficiency in routing event messages while enforcing privacy by increasing the number of rules that can be assigned to a network device. In particular, the thesis will propose an architecture that enables the enforcement of caching of flows in a scalable manner and will design and compare different cache replacement strategies that optimize for average end-to-end latency in forwarding events between publishers and subscribers. The thesis will evaluate the efficiency of the proposed model as part of a prototypical implementation and document the findings in written form.

Literature:

• Sukanya Bhowmik, Muhammad Adnan Tariq, Boris Koldehofe, Frank Dürr, Thomas Kohler, and Kurt Rothermel. High Performance Publish/Subscribe Middleware in Software-Defined Networks.  IEEE/ACM Transactions on Networking, 25(3):1501--1516, june 2017. https://doi.org/10.1109/TNET.2016.2632970

• Jannik Lange. Privacy-Preserving Content-based Routing in Software-Defined Networks. [Abstract]

Suitability for student:

Master

Contact:

ahmad.baghdadi [at] uni-marburg.de