Dr. André Martin (TU Dresden)
1. Introduction into ESP / Dependability
In the first lecture of the dependability course, we will introduce you to the fundamentals of Event Stream Processing (ESP) systems such as the data, query and execution model. In order to get familiar with ESP systems, we will then take a look at a couple of real world streaming applications and their implementations originating from the annual DEBS challenge. The two application we will look at cover the following use cases: energy consumption predication as well as an online analysis of taxi rides in NYC area. We will then provide an overview about dependability fundamentals such as common failures models, recover guarantees etc. tailored to ESP systems. The lecture closes with an overview about common fault tolerance mechanisms used in ESP systems.
2. Passive Replication
In the second lecture of the dependability class, we will take a closer look at recovery protocols used in passive replication. After getting familiar with those approaches, we will discuss challenges with passive replication such as the introduced overhead due to check-pointing and event ordering. Several approaches will be presented in order to reduce the overhead of check-pointing and event ordering based on latest research.
3. Active Replication
The third lecture in this course is dedicated to active replication, an alternative to passive replication. We will first introduce you to the principles of active replication and discuss challenges such as how to achieve consistency across replicas. You will learn about several different protocols to achieve consistency using a deterministic merge. Furthermore, we discuss how the overhead can be reduced for several classes of applications using slightly weaker consistency semantics.
4. Adaptive Fault Tolerance
The last lecture of the course covers an adaptive approach where several replication strategies are combined and adapted during run-time. You will learn about two different approaches tailored to cloud environments that allow us to reduce the resource overhead of active replication and improve availability at the same time.
Back to EBSIS Events section.