Simon Kiertscher

Cherub: Power Consumption Aware Cluster Resource Management

This paper presents the design and implementation of an energy saving daemon for clusters called cherub. The design of the cherub daemon is modular and extensible. In the field of High Performance Computing (HPC) well known Resource Management Systems (RMSs) like Portable Batch System (PBS) [8]or its open source derivative, the TORQUE resource manager [11]are used to manage clusters and work queues. Thus, cherub is able to interact with different RMSs to make them energy aware. cherub is also suited for load-balancing clusters managed by dispatchers like Linux Virtual Server (LVS) [17] since the only requirement is a central approach for resource management.

IPv6 Network Attack Detection with HoneydV6

During 2012, we conducted a long term IPv6-darknet experiment. We observed a relatively high number of interesting events and therefore needed additional network security tools to capture and analyse potentially harmful IPv6 traffic. This paper presents HoneydV6, a low-interaction IPv6 honeypot that can simulate entire IPv6 networks and which may be utilized to detect and analyze IPv6 network attacks. Our implementation is based on the well-known low-interaction honeypot Honeyd. To the best of our knowledge, this is the first low-interaction honeypot which is able to simulate entire IPv6 networks on a single host. Enticing attackers to exploit an IPv6 honeypot requires new approaches and concepts because of the huge IPv6 address space. We solved this problem through a dynamic instantiation mechanism that increases the likelihood for an attacker to find a target host in our IPv6 honeynet.

Energy Aware Resource Management for Clusters of Web Servers

Web server clusters guarantee high performance and high availability for applications like google and amazon. Typically, over-provisioning is used to guarantee a Service Level Agreement in the case of peak load situations. We present an energy saving mechanism which extends the capabilities of traditional load balancers like for example the Linux Virtual Server (LVS). The energy saving daemon called Cherub, running on the front node of the cluster, turns nodes on and off depending on the current load situation. We evaluate different load determination and forecasting methods to detect the utilization of the back end servers in a test bed using requests from a web server log file of the Wikimedia Foundation. Even for the most challenging situation, i.e. in case of a peak load, LVS enhanced with Cherub saves 92% of the theoretical possible optimum saving, while keeping the service level agreement at 99.4% in the experiment.

Scalability evaluation of an energy-aware resource management system for clusters of web servers

For green cluster computing resource management systems have to be energy-aware. CHERUB is such an energy-aware resource management system which works together with the Linux Virtual Server. Experiments in a small cluster setup with two nodes have shown the benefit of CHERUB. This paper presents necessary design changes to make CHERUB also work in big cluster setups. Our methodological approach is two-fold. First, we present unit measurements to evaluate the scaling of the re-implemented functions. Second, a cluster simulator is presented and validated which makes it possible to test CHERUB for backend clusters of arbitrary size.

Evaluation of Threshold-based Fall Detection on Android Smartphones

This paper evaluates threshold-based fall detection algorithms which use data from acceleration sensors that are part of the current smartphone technology. Different detection algorithms are published in the literature with different threshold values. This paper presents the evaluation of 5 different algorithms which are suited for Android smartphones. In contradiction to prior work, our experiments indicate that the Free Fall detection Phase is necessary for a low False Positive Rate. Further, we present an empirical evaluation of currently available fall detection apps in the Google Play store.

Optimizing Energy Efficiency and Quality of Service in Large Scale Web Server Environments

Energy-aware resource-management strategies for web-server clusters use on/off strategies which are based on thresholds. The basic idea is to turn off currently unused nodes, and on again when the load increases. The biggest challenge for these strategies is to provide enough compute power in an unexpected peak load situation. We investigate this challenge using trace-driven simulation for different scenarios like using Suspend-to-RAM capable nodes, over-provisioning, load forecasting, and anti-flapping strategies. We evaluate the results under different new score metrics which represent different cluster operating modes (high performance, low energy costs, balanced) and determine the corresponding optimum strategies. For example, for the balanced scenario, the proposed on/off strategy achieves 28.08 % energy saving while retaining a quality of service of 96.99 % even in the peak scenario. In contrast, for the same scenario the optimum strategy achieves an energy saving of 29.58 %.

Self-Adapting Load Balancing for DNS

The Domain Name System belongs to the core services of the Internet infrastructure. Hence, DNS availability and performance is essential for the operation of the Internet and replication and load balancing is used for the root and top level name servers. This paper proposes an architecture for credit based server load balancing (SLB) for DNS. Compared to traditional load balancing algorithms like round robin or least connection, the benefit of credit based SLB is that the load balancer can adapt more easily to heterogeneous load requests and heterogeneous back end server capacities. The challenge of this approach is the definition of a suited credit metric. While this was done before for TCP based services like HTTP, the problem was not solved for UDP based services like DNS. This paper presents an approach to define credits also for UDP based services. This UDP/DNS approach is implemented within the credit based SLB implementation salbnet. The presented measurements confirm the benefit of the self-adapting credit based SLB approach.

Self-adapting load balancing for DNS

The Domain Name System belongs to the core services of the Internet infrastructure. Hence, DNS availability and performance is essential for the operation of the Internet and replication and load balancing is used for the root and top level name servers. This paper proposes an architecture for credit based server load balancing (SLB) for DNS. Compared to traditional load balancing algorithms like round robin or least connection, the benefit of credit based SLB is that the load balancer can adapt more easily to heterogeneous load requests and heterogeneous back end server capacities. The challenge of this approach is the definition of a suited credit metric. While this was done before for TCP based services like HTTP, the problem was not solved for UDP based services like DNS. This paper presents an approach to define credits also for UDP based services. This UDP/DNS approach is implemented within the credit based SLB implementation salbnet. The presented measurements confirm the benefit of the self-adapting credit based SLB approach.