Linh Thi Xuan Phan

Realizing Compositional Scheduling through Virtualization

We present a co-designed scheduling framework and platform architecture that together support compositional scheduling of real-time systems. The architecture is built on the Xen virtualization platform, and relies on compositional scheduling theory that uses periodic resource models as component interfaces. We implement resource models as periodic servers and consider enhancements to periodic server design that significantly improve response times of tasks and resource utilization in the system while preserving theoretical schedulability results. We present an extensive evaluation of our implementation using workloads from an avionics case study as well as synthetic ones.

Real-time multi-core virtual machine scheduling in xen

Recent years have witnessed two major trends in the development of complex real-time embedded systems. First, to reduce cost and enhance flexibility, multiple systems are sharing common computing platforms via virtualization technology, instead of being deployed separately on physically isolated hosts. Second, multicore processors are increasingly being used in real-time systems. The integration of real-time systems as virtual machines (VMs) atop common multicore platforms raises significant new research challenges in meeting the real-time performance requirements of multiple systems. This paper advances the state of the art in real-time virtualization by designing and implementing RT-Xen 2.0, a new real-time multicore VM scheduling framework in the popular Xen virtual machine monitor (VMM). RT-Xen 2.0 realizes a suite of real-time VM scheduling policies spanning the design space. We implement both global and partitioned VM schedulers; each scheduler can be configured to support dynamic or static priorities and to run VMs as periodic or deferrable servers. We present a comprehensive experimental evaluation that provides important insights into real-time scheduling on virtualized multicore platforms: (1) both global and partitioned VM scheduling can be implemented in the VMM at moderate overhead; (2) at the VMM level, while compositional scheduling theory shows partitioned EDF (pEDF) is better than global EDF (gEDF) in providing schedulability guarantees, in our experiments their performance is reversed in terms of the fraction of workloads that meet their deadlines on virtualized multicore platforms; (3) at the guest OS level, pEDF requests a smaller total VCPU bandwidth than gEDF based on compositional scheduling analysis, and therefore using pEDF at the guest OS level leads to more schedulable workloads in our experiments; (4) a combination of pEDF in the guest OS and gEDF in the VMM - configured with deferrable server - leads to the highest fraction of schedulable task sets compared to other real-time VM scheduling policies; and (5) on a platform with a shared last-level cache, the benefits of global scheduling outweigh the cache penalty incurred by VM migration.

Composing Functional and State-Based Performance Models for Analyzing Heterogeneous Real-Time Systems

We present a performance analysis technique for distributed real-time systems in a setting where certain components are modeled in a purely functional manner, while the remaining components require additional modeling of state information. The functional models can be efficiently analyzed but have restricted expressiveness. On the other hand, state-based models are more expressive and offer a richer set of analyzable properties but are computationally more expensive to analyze. We show that by appropriately composing these two classes of models it is possible to leverage on their respective advantages. To this end, we propose an interface between components that are modeled using real-time calculus [Chakraborty, Kiinzli and Thiele, DATE 2003] and those that are modeled using event count automata [Chakraborty, Phan and Thiagarajan, RTSS 2005]. The resulting modeling technique is as expressive as event count automata, but is amenable to more efficient analysis. We illustrate these advantages using a number of examples and a detailed case study.

Compositional Analysis of Multi-mode Systems

The paper presents a model for multi-mode real-time applications and develops new techniques for the compositional analysis of systems that contain multiple such applications. An algorithm for constructing an interface for a single multi-mode application is presented. Then, a method for computing an interface of a composite application is presented, which uses only the interfaces of constituent applications. A case study of an adaptive streaming system demonstrates that multi-mode analysis offers more precise results compared to a unimodal worst-case analysis.

Event count automata: a state-based model for stream processing systems

Recently there has been a growing interest in models and methods targeted towards the (co)design of stream processing applications; e.g. those for audio/video processing. Streams processed by such applications tend to be highly bursty and exhibit a high data-dependent variability in their processing requirements. As a result, classical event and service models such as periodic, sporadic, etc. can be overly pessimistic when dealing with such applications. In this paper, we present a new model called event count automata (ECA) for capturing the timing properties of such streams. Our model can be used to cleanly formulate properties relevant to stream processing on heterogeneous multiprocessor architectures, such as buffer overflow/underflow constraints. It can also provide the basis for developing analysis methods to compute delay/timing properties of the processed streams under different scheduling policies. Our ECAs, though similar in flavor to timed and hybrid automata, have a different semantics, are more light-weight, and are specifically suited for modeling stream processing applications and architectures. We present the basic aspects of this model and illustrate its modeling potential. We then apply it in a specific stream processing setting and develop an analysis technique based on the formalism of colored Petri nets (CPNs). Finally, we validate our modeling and analysis techniques with the help of preliminary experimental results generated using the CPN simulation tool

An empirical analysis of scheduling techniques for real-time cloud-based data processing

In this paper, we explore the challenges and needs of current cloud infrastructures, to better support cloud-based data-intensive applications that are not only latency-sensitive but also require strong timing guarantees. These applications have strict deadlines (e.g., to perform time-dependent mission critical tasks or to complete real-time control decisions using a human-in-the-loop), and deadline misses are undesirable. To highlight the challenges in this space, we provide a case study of the online scheduling of MapReduce jobs executed by Hadoop. Our evaluations on Amazon EC2 show that the existing Hadoop scheduler is ill-equipped to handle jobs with deadlines. However, by adapting existing multiprocessor scheduling techniques for the cloud environment, we observe significant performance improvements in minimizing missed deadlines and tardiness. Based on our case study, we discuss a range of challenges in this domain posed by virtualization and scale, and propose our research agenda centered around the application of advanced real-time scheduling techniques in the cloud environment.

Towards dynamic pricing-based collaborative optimizations for green data centers

Increased demand for cloud computing services has ushered power management schemes into the frontlines of data center research. Meanwhile, market penetration of intermittent renewable energy sources (e.g., wind and solar) is on the rise. While clean and abundant, their intermittency is troubling for utility companies, requiring power balancing reserves to be deployed at anytime to precisely match consumer demand with energy availability. However, a transformative redesign of our power grid is looming, calling for the use of dynamic energy pricing to resolve this issue by possibly shaping demand. Data centers, being significant consumers with the ability to adjust power utilization in real-time (e.g., by migrating its jobs to and from other locations), are ideal candidates to participate in dynamic pricing markets. We propose a collaborative cost optimization framework by coupling utilities with data centers via dynamic pricing. We develop models describing the information exchange framework for utilities and data centers and employ a distributed constraint optimization solver, Cologne, to negotiate a mutually optimal price. An evaluation of our system has been performed using real intermittent-energy-generation trace data. Modeling the dynamic price over this trace, we show that our technique could reduce a participating data centers costs by 75%. On the side of utilities, we further show that consumer power demand can be shaped to reveal a 17% improvement on average.

Network functions virtualization with soft real-time guarantees

Network functions are increasingly being commoditized as software appliances on off-the-shelf machines, popularly known as Network Functions Virtualization (NFV). While this trend provides economics of scale, a key challenge is to ensure that the performance of virtual appliances match that of hardware boxes. We present the design and implementation of NFV-RT, a system that dynamically provisions resources in an NFV environment to provide timing guarantees. Specifically, given a set of service chains that each consist of some network functions, NFV-RT aims at maximizing the total number of requests that can be assigned to the cloud for each service chain, while ensuring that the assigned requests meet their deadlines. Our approach uses a linear programming model with randomized rounding to efficiently and proactively obtain a near-optimal solution. Our simulation shows that, given a cloud with thousands of machines and service chains, NFV-RT requires only a few seconds to compute the solution, while accepting three times the requests compared to baseline heuristics. In addition, under some special settings, NFV-RT can provide significant performance improvement. Our evaluation on a local testbed shows that 94% of the packets of the submitted requests meet their deadlines, which is three times that of previous reactive-based solutions.

CARTS: a tool for compositional analysis of real-time systems

As real-time embedded systems are increasingly complex, integration becomes a great challenge in their design and development. Managing complexity of the system design is therefore essential for high-assurance and cost-effective development. Component-based design has consequently been developed and gained its importance over the years as a powerful technique for complexity management. In this design paradigm, a large complex system is first decomposed into smaller and simpler components - which are developed independently - before recomposing them into a complete system using interfaces that abstract away their internal complexities.

Co-design of control and platform with dropped signals

This paper examines a co-design of control and platform in the presence of dropped signals. In a cyber-physical system, due to increasing complexities such as the simultaneous control of several applications, limited resources, and complex platform architectures, some of the signals transmitted may often be dropped. In this paper, we address the challenges that arise both from the control design and the platform design point of view. A dynamic model is proposed that accommodates these drops, and a suitable switching control design is proposed. A Multiple Lyapunov functions based approach is used to guarantee the stability of the system with the switching controller. We then present a method for optimizing the amount of platform resource required to ensure stability of the control systems via a buffer control mechanism that exploits the ability to drop signals of the control system and an associated analysis of the drop bound. The results are demonstrated using a case study of a co-designed lane keeping control system in the presence of dropped signals.