- ParPerOS: Parallel Persistency OS (DFG: LO 1719/8-1 and and DI 2840/2-1)
- In ParPerOS, we examine new abstractions for unified but efficient and optionally crash-consistent low-level memory management for data objects in heterogeneous memory systems that consist of volatile, persistent, distributed and other types of main memory.
Our former SRA member and current project partner in the ATLAS and ParPerOS projects, Christian Dietrich (TUHH) helds his inaugural lecture on New Directions for Managing Memory:
Abstract: Traditionally, memory is the scarce resource that operating systems virtualize for their users. However, current hardware trends, like ultra-fast NVMe SSDs and non-volatile RAM, force us to rethink operating system-mediated management. We no longer have to manage scarcity, but we have to swim in the new abundance without drowning. In his inaugural lecture, Christian Dietrich will present three ongoing research projects that center around the topic of memory management.
The event starts at 14:00 and can be followed by Zoom.
- ATLAS: Adaptable Thread-Level Address Spaces (DFG: LO 1719/7-1 and DI 2840/1-1)
- In the ATLAS project, we investigate dynamic specialization and containment by means of thread-level address-space variations.
Abstract: Computer-based automation in industrial appliances led to a growing number of logically dependent, but physically separated embedded control units per appliance. Many of those components are safety-critical systems, and require adherence to safety standards, which is inconsonant with the relentless demand for features in those appliances. Features lead to a growing amount of control units per appliance, and to a increasing complexity of the overall software stack, being unfavourable for safety certifications. Modern CPUs provide means to revise traditional separa- tion of concerns design primitives: the consolidation of systems, which yields new engineering challenges that concern the entire software and system stack.
Multi-core CPUs favour economic consolidation of formerly separated systems with one efficient single hardware unit. Nonetheless, the system architecture must provide means to guarantee the freedom from interference between domains of different criticality. System consolidation demands for architectural and engineering strategies to fulfil requirements (e.g., real-time or certifiability criteria) in safety-critical environments.
In parallel, there is an ongoing trend to substitute ordinary proprietary base platform software components by mature OSS variants for economic and engineering reasons. There are funda- mental differences of processual properties in development processes of OSS and proprietary software. OSS in safety-critical systems requires development process assessment techniques to build an evidence-based fundament for certification efforts that is based upon empirical software engineering methods.
In this thesis, I will approach from both sides: the software and system engineering perspective. In the first part of this thesis, I focus on the assessment of OSS components: I develop software engineering techniques that allow to quantify characteristics of distributed OSS development processes. I show that ex-post analyses of software development processes can be used to serve as a foundation for certification efforts, as it is required for safety-critical systems.
In the second part of this thesis, I present a system architecture based on OSS components that allows for consolidation of mixed-criticality systems on a single platform. Therefore, I exploit virtualisation extensions of modern CPUs to strictly isolate domains of different criticality. The proposed architecture shall eradicate any remaining hypervisor activity in order to preserve real- time capabilities of the hardware by design, while guaranteeing strict isolation across domains.
- CADOS: Configurability-Aware Development of Operating Systems (DFG: LO 1719/3-2)
- In the CADOS project, we investigate scalable methods and tools to deal with the implementation of variability across all implementation layers of modern system software.
After many fruitful years with dozen of papers, great lectures and a lot of fun together, Christian Dietrich leaves our group to start his Juniorprofessorship (W1-TT-W3) with a new operating system group at TUHH. We will continue our work together, nevertheless miss him a lot, and wholeheartedly congratulate Prof. Dr.-Ing. Christian Dietrich for this great step in his career!
Christian Dietrich receives an award for the best doctoral thesis in the field of operating systems. The award is granted annually by the SIG on Operating Systems of the German Computer Assiciation (GI Fachgruppe Betriebssysteme) solely on the base of scientific excellence. It includes a price money of 500 €. Congrats, Christian!
In his dissertation Interaction-Aware Analysis and Optimization of Real-Time Application and Operating System, Christian designs and implements a control-flow--sensitive whole-system view and analysis on the interactions within real-time systems. With this approach, he can overcome many inefficiencies that arise from analyses that have an isolating focus on individual system components. Furthermore, the interaction-aware methods keep close to the actual implementation, and therefore are able to consider the behavioral patterns of the finally deployed real-time computing system.
Florian Rommel presents our paper From Global to Local Quiescence: Wait-Free Code Patching of Multi-Threaded Processes at OSDI '20 – due to Corona by video.
In the paper, we present WfPatch, a wait-free approach to inject code changes into running multi-threaded programs. Instead of having to stop the world before applying a patch, WfPatch can gradually apply it to each thread individually at a local point of quiescence, while all other threads can make uninterrupted progress.
WfPatch is the first outcome of our novel concept on adaptable thread-level address spaces, which we are investigating in the ATLAS project.
Tobias Landsberg receives the award for best master thesis in the field of operating systems. The award is granted annually by the SIG on Operating Systems of the German Computer Assiciation (GI Fachgruppe Betriebssysteme) solely on the base of scientific excellence. It includes a price money of 500 €. Congrats, Tobias!
In his thesis Analyzing and Optimizing TLB-Induced Thread Migration Costs on Linux/ARM Tobias evaluates if it is feasable and benefitial to pre-warm the TLB (Translation Look-Aside Buffer) when a thread is migrated to another core. He analyzes existing ARM cores, presents and evaluates possible hardware extensions in gem5 and and provides a complete Linux integration for the system.
Abstract: Mechanical and electronic automation was a key component of the technological advances in the last two hundred years. With the use of special-purpose machines, manual labor was replaced by mechanical motion, leaving workers with the operation of these machines, before also this task was conquered by embedded control systems. With the advances of general-purpose computing, the development of these control systems shifted more and more from a problem-specific one to a one-size-fits-all mentality as the trade-off between per-instance overheads and development costs was in favor of flexible and reusable implementations. However, with a scaling factor of thousands, if not millions, of deployed devices, overheads and inefficiencies accumulate; calling for a higher degree of specialization.
For the area of real-time operating systems, which form the base layer for many of these computerized control systems, we deploy way more flexibility than what is actually required for the applications that run on top of it. Since only the solution, but not the problem, became less specific to the control problem at hand, we have the chance to cut away inefficiencies, improve on system-analyses results, and optimize the resource consumption. However, such a tailoring will only be favorable if it can be performed without much developer interaction and in an automated fashion. Here, real-time systems are a good starting point, since we already have to have a large degree of static knowledge in order to guarantee their timeliness. Until now, this static nature is not exploited to its full extent and optimization potentials are left unused.
The requirements of a system, with regard to the RTOS, manifest in the interactions between the application and the kernel. Threads request resources from the RTOS, which in return determines and enforces a scheduling order that will ensure the timely completion of all necessary computations. Since the RTOS runs only in the exception, its reaction to requests from the application (or from the environment) is its defining feature.
In this thesis, I will grasp these interactions, and thereby the required RTOS semantic, in a control-flow--sensitive fashion. Extracted automatically, this knowledge about the reciprocal influence allows me to fit the implementation of a system closer to its actual requirements. The result is a system that is not only in its usage a special-purpose system, but also in its implementation and in its provided guarantees.
In the development of my approach, it became clear that the focus on these interactions is not only highly fruitful for the optimization of a system, but also for its end-to-end analysis. Therefore, this thesis does not only provide methods to reduce the kernel-execution overhead and a system's memory consumption, but it also includes methods to calculate tighter response-time bounds and to give guarantees about the correct behavior of the kernel. All these contributions are enabled by my proposed interaction-aware methodology that takes the whole system, RTOS and application, into account.
With this thesis, I show that a control-flow--sensitive whole-system view on the interactions is feasible and highly rewarding. With this approach, we can overcome many inefficiencies that arise from analyses that have an isolating focus on individual system components. Furthermore, the interaction-aware methods keep close to the actual implementation, and therefore are able to consider the behavioral patterns of the finally deployed real-time computing system.
Abstract: In der heutigen Wissenschaft und Wirtschaft haben wir es oft mit Systemen zu tun, welche aus Problemen bestehen, die sehr komplex und nicht einfach zu lösen sind. Aufgrund der zunehmenden Komplexität und der teilweise fehlenden Informationen ist es bereits heutzutage nicht mehr möglich, solche Probleme – welche als Blackbox-Probleme klassifiziert werden – per Hand zu lösen. Um das Maximum oder Minimum zu finden, wird auf Optimierungsmethoden zurückgegriffen, die uns ermöglichen, eine optimale Lösung für das Problem zu suchen und ggf. zu finden. Stochastische Methoden haben die letzten Jahre gezeigt, dass sie sehr gut geeignet sind, solche Probleme zu lösen. Der Vorteil der Verwendung von stochastischen Methoden ist, dass sie nicht den Gradienten des zu optimierenden Problems verwenden, so dass sie sowohl bei großen als auch bei komplexen Optimierungsproblemen erfolgreich angewendet werden können. Diese Vielseitigkeit hat aber ihren Preis. Es gibt hauptsächlich drei wesentliche Aspekte, die die Effizienz der Lösung beeinträchtigen:
- Die realen Probleme werden immer größer und komplizierter oder sie müssen in sehr kurzer Zeit gelöst werden, was erhebliche Ressourcen in Zeit und Hardware erfordert.
- Optimierungsprobleme sind durch mehrere lokale Optima charakterisiert, die ein Verfahren zur Vermeidung einer zu frühen Konvergenz erfordern.
- Algorithmen erfordern einige problembedingte Anpassungen ihrer Verhaltensparameter, um bessere Ergebnisse zu erzielen.
Untersuchungen in dieser Arbeit haben gezeigt, dass die Anpassungen zu besse ren Ergebnissen führen. Durch die adaptive Natur des Frameworks, ist es in vielen Rechnerarchitekturen nutzbar und für viele Probleme anwendbar.
Prof. Dr.-Ing. habil. Daniel Lohmann gave his inaugural lecture at the Faculty of Electrical Engineering and Computer Science. In his presentation "Klein und sicher – Automatisch anpassbare Systemsoftware für eingebettete Spezialzweckanwendungen", Prof. Lohmann provided an entertaining introduction into our research activities and the case for highly tailorable system software.
Björn Fiedler presents our paper Levels of Specialization in Real-Time Operating Systems was at the 14th Workshop on Operating System Platforms for Embedded Real-Time Applications (OSPERT '18), in Barcelona. In the paper we describe a taxonomy for the specialization of system software towards a specific application and provide showcases of the achievable benefits. We got an Best Paper Award for this work.
- AHA: Automated Hardware Abstraction in Operating-System Engineering (DFG: LO 1719/4-1)
- Goal of AHA is to improve nonfunctional properties of system software by a very deep, but fully automated specialization of the application-hardware bridge represented by the operating system. We investigate, how alternative implementations that are mapped more directly to hardware features, can be generated from a concrete application and their actual interactions with the operating system.
Die Kriterien für die Auszeichnung sind eine herausragende Lehrleistung über die Dauer von wenigstens zwei Studienjahren an einer Universität in Bayern, eine Beteiligung der Studierenden an der Auswahl sowie der Vorschlag der jeweiligen Universität. Über alle Maßnahmen zur Sicherung der Qualität der Lehre, die von den Hochschulen praktiziert werden, spielen das persönliche Engagement und die pädagogisch-didaktischen Kompetenzen des Lehrenden eine große Rolle.