Browsing by Author "Kutzleb, André"

Now showing 1 - 2 of 2

Open Access
Integrating storage components to control event-rates in SDN-based content routing networks
(2014) Kutzleb, André
Software-defined Networking (SDN) makes it possible to build content-based routing systems which achieve line-rate performance. As recent work has shown, a content-based publish/subscribe system can be built using SDN-switches instead of software-based brokers found in traditional publish/subscribe systems to forward events. These switches provide content-based routing just like the brokers in traditional publish/subscribe systems, however, they are operating on hardware and simply forward events according to their forwarding tables. While this eliminates the need for a network of software-based brokers and results in a significant performance boost, it also requires solutions for problems not existing or already solved in broker-based publish/subscribe systems. One of these problems is the control of event rates. While content producers (publishers) in traditional publish/subscribe systems depend on brokers to forward their events, in a SDN-based system a publisher can simply publish at line-rate with no broker as intermediary to control the event rate - the switches will simply forward any event to the content consumers (subscribers) at a rate limited only by the forwarding capability of their hardware. Because of this lack of any kind of rate control, an additional mechanism is necessary in order to avoid the overload of subscribers. This thesis provides a solution to this problem by placing event caches in the network which temporarily buffer events whenever a subscriber is overloaded and retransmit them at a controlled rate once it is possible to do so without overloading the subscriber. Additionally, an algorithm for finding good positions in the network to place caches at is proposed. Evaluations of this caching mechanism show the feasibility of this approach for controlling event rates. These evaluations show that in some situations a reduction in event loss of more than 99.7\% has been achieved with the introduction of caches.
Open Access
Visual analytics of big data from distributed systems
(2017) Kutzleb, André
Distributed Systems are challenging to debug because the temporal order of events and distributed states are hard to track. The high complexity of distributed systems make fully automatic reasoning difficult to apply. Domain experts are often required to reason about the behavior of a system based on log files from various sources. This situation presents a good opportunity for visual analytics. Data from multiple sources can be preprocessed and visualized, and then domain experts can conduct exploratory analysis to accelerate the identification of issues. The goal of this master thesis was to create such a visual analytics tool to help domain experts explore data collected from distributed systems more efficiently and assist in identifying bugs and anomalies. The system was used by domain experts and helped to identify issues in a distributed system, showing that visual analytics can be a useful tool to assist domain experts in their daily work.