The number of distributed applications that play important roles in industry, commerce, and daily life is steadily increasing. The execution behavior constraints that distributed applications must meet vary widely, but those of the important sub-class, the distributed control loops, are the focus of the work described in this report. Distributed control loops have two characteristics of particular interest: (1) components of the application communicate with each other across machine boundaries, and (2) the end-toend response time and other aspects of control loop behavior are subject to specified timing constraints. Distributed control loops have been implemented for decades, but generally using specialized computation platforms. Recent trends make supporting such control loops alongside other applications on low cost commercial off the shelf (COTS) platforms, particularly open source platforms, increasingly attractive. The viability of this approach depends crucially on which aspects of these low cost platforms constrain control loop performance, what the constraints are, and where in the system they are created. This report describes a number of experiments which explore the performance envelope of control loops on Linux using the increasingly popular RT-Patch, and which

Abstract—Nowadays, high computing demands are often tackled by clusters of single computers, each of which is basically an assembly of a growing number of CPU cores and main memory, also called a node; these nodes are connected by some kind of communication network. With the growing speed and number of CPU cores, the network becomes a severe bottleneck limiting overall cluster performance. Highspeed interconnects like InfiniBand, SCI, and Dolphin DX are good for alleviating this communication bottleneck, when the communication software is either based on IP or specifically adapted to the interconnect. Software written to communicate directly via Ethernet frames can not be used this way, though. In this article, we present two drivers for Linux that fill this gap. ETHOS is a very generic Ethernet over Sockets driver. With this driver it is possible to use any interconnect that offers a sockets interface as replacement for Ethernet. The second driver, ETHOM, sacrifices the compatibility with a wide range of interconnects in favour of higher performance on top of Dolphin’s high-speed networks SCI and DX. It enhances their functionality by offering an Ethernet and with that an IP interface. Both drivers allow usage of layer-2 kernel functionality like interface bonding and bridging. By means of various measurements, we show that ETHOS and ETHOM with InfiniBand, SCI, or DX offer a two- to threefold increase in communication performance over Gigabit Ethernet.

Abstract: Snort is the most widely deployed network intrusion detection system (NIDS) worldwide, with millions of downloads to date. PC-based Snort typically runs on either Linux or Windows operating systems. In this paper, we present an experimental evaluation and comparison of the performance of Snort NIDS when running under the two newly released operating systems of Windows 7 and Windows Server 2008. Snort's performance is measured when subjecting a PC host running Snort to both normal and malicious traffic. Snort's performance is evaluated and compared in terms of throughput and packet loss. In order to offer sound interpretations and get a better insight into the behaviour of Snort, we also measure the packet loss encountered at the kernel level. In addition, we study the impact of running Snort under different system configurations which include CPU scheduling priority given to user applications or kernel services, uni and multiprocessor environment, and processor affinity.