Pratyush Kumar

Evaluating example-based search tools

A crucial element in consumer electronic commerce is a catalog tool that not only finds the product for the user, but also convinces him that he has made the best choice. To do that, it is important to show him ample choices while keeping his interaction effort below an acceptable limit. Among the various interaction models used in operational e-commerce sites, ranked lists are by far the most popular tool for product navigation and selection. However, as the number of product features and the complexity of users criteria increase, a ranked lists efficiency becomes less satisfactory. As an alternative, research groups from the intelligent user interface community have developed various example-based search tools, including SmartClient from our laboratory. These tools not only perform personalized search, but also support tradeoff analysis. However, despite the academic interest, example-based search paradigms have not been widely adopted in practice. We have examined the usability of such tools on a variety of tasks involving selection and tradeoff. The studies clearly show that example-based search is comparable to ranked lists on simple tasks, but significantly reduces the error rate and search time when complex tradeoffs are involved. This shows that such tools are likely to be useful particularly for extending the scope of consumer e-commerce to more complex products.

A hybrid approach to cyber-physical systems verification

We propose a performance verification technique for cyber-physical systems that consist of multiple control loops implemented on a distributed architecture. The architectures we consider are fairly generic and arise in domains such as automotive and industrial automation; they are multiple processors or electronic control units (ECUs) communicating over buses like FlexRay and CAN. Current practice involves analyzing the architecture to estimate worst-case end-to-end message delays and using these delays to design the control applications. This involves a significant amount of pessimism since the worst-case delays often occur very rarely. We show how to combine functional analysis techniques with model checking in order to derive a delay-frequency interface that quantifies the interleavings between messages with worst-case delays and those with smaller delays. In other words, we bound the frequency with which control messages might suffer the worst-case delay. We show that such a delay-frequency interface enables us to verify much tigher control performance properties compared to what would be possible with only worst-case delay bounds.

Thermally optimal stop-go scheduling of task graphs with real-time constraints

Dynamic thermal management (DTM) techniques to manage the load on a system to avoid thermal hazards are soon becoming mainstream in todays systems. With the increasing percentage of leakage power, switching off the processors is becoming a viable alternative technique to speed scaling. For real-time applications, it is crucial that under such techniques the system still meets the performance constraints. In this paper we study stop-go scheduling to minimize peak temperature when scheduling an application, modeled as a task-graph, within a given makespan constraint. For a given static-ordering of execution of the tasks, we derive the optimal schedule referred to as the JUST schedule. We prove that for periodic task-graphs, the optimal temperature is independent of the chosen static-ordering when following the proposed JUST schedule. Simulation experiments validate the theoretical results.

Cool shapers: shaping real-time tasks for improved thermal guarantees

With increasing power densities, managing on-chip temperatures has become an important design challenge. We propose a novel approach to this problem with the use of shapers to dynamically and selectively insert idle times during the execution of hard real-time jobs on a single speed processor. For the class of leaky bucket shapers which have a light-weight implementation, we derive the shaper such that no job misses its real-time deadline and the peak temperature is optimally reduced. The analysis and design of such shapers allows for dynamically variable streams of jobs; for instance, periodic streams with jitter. We extend our results to consider non-zero power and timing overhead in transitioning to the idle mode. With experimental results, we demonstrate that the proposed approach provides a large improvement: on average 8K peak temperature reduction or 40% increase in utilization for a given peak temperature.

Energy efficient DVFS scheduling for mixed-criticality systems

Consolidating functionalities with different safety requirements into a common platform gives rise to mixed-criticality systems. The state-of-the-art research has focused on providing heterogeneous timing guarantees for tasks of varying criticality levels. This is achieved by dropping less critical tasks when critical tasks overrun. However, with drastically increased computing requirements and the often battery-operated nature of mixed-criticality systems, energy minimization for such systems is also becoming crucial. In fact, this has already been possible since many modern processors are equipped with the capacity of dynamic voltage and frequency scaling (DVFS), where processor frequency can be reduced at runtime to save energy. We present in this paper the first results known to date on applying DVFS to mixed-criticality systems. We show that DVFS can be used to help critical tasks to meet deadlines by speeding up the processor when they overrun. This will further allow the system to reserve less time budgets for task overrun. Thus, more slack can be explored to reduce the processor frequency to save energy for scenarios when tasks do not overrun. Since overrun is rare, such a strategy can greatly reduce the expected energy consumption for mixed-criticality systems. For solving the energy minimization problem, we formulate a convex program by integrating DVFS with a well-known mixed-criticality scheduling technique - EDF-VD. Furthermore, we present analytical results on this problem and propose an optimal algorithm to solve it. With both theoretical and experimental results, we demonstrate energy savings and various tradeoffs.

Interference Constraint Graph — A new specification for mixed-criticality systems

Current research in mixed-criticality systems assumes that any task of lower criticality levels can be dropped at anytime in order to guarantee the schedulability of tasks of higher criticality levels. However, in an industrial mixed-criticality system, tasks may interfere with each other only under certain scenarios. Currently a designer does not have any means to specify or control this. The paper proposes the Interference Constraint Graph (ICG) which specifies the allowed interferences between tasks. The new specification formalism generalizes and can easily express many of the existing mixed-criticality scheduling conditions. In spite of its generality, we show that standard fixed-priority scheduling can be efficiently applied. Experiments demonstrate that the ICG model enables systematic reduction of the number of tasks that can be dropped.

Timing Analysis on a Processor with Temperature-Controlled Speed Scaling

Several recent works consider the problem of temperature-constrained scheduling of jobs. In such attempts, speed of the processor and the execution of jobs is software-controlled such that temperature and performance constraints are met. An alternative approach is to use measurements from temperature sensors to actuate the speed of the processor as a feedback control loop. Though such a solution explicitly and independently meets the thermal constraints, the analysis of the real-time properties of tasks served by such a processor is not straightforward. In this paper, we study this problem for a variable stream of jobs characterized by an input arrival rate. We show that an intuitive notion of monotonicity extends to such a processor. Using this property, we present an analytical technique to determine the worst-case delay suffered by jobs. The presented technique efficiently and tightly determines the delay as a function of the initial temperature. The simplicity of this analysis motivates further analysis and mainstream use of such systems.

Quantifying the Effect of Rare Timing Events with Settling-Time and Overshoot

For hard real-time systems, worst-case timing models are employed to validate whether timeliness properties, such as meeting deadlines, are always satisfied. We argue that such a deadline-interface should be generalised in view of two separate motivations: (a) applications can tolerate bounded non-satisfaction of timeliness properties due to inherent robustness or relaxed quality requirements, and (b) worst-case timing models do not expose the occurrence of certain rare yet predictable events. As a more expressive interface, we propose the Rare-Event with Settling-Time (REST) model wherein, during rare events nominal timing models can be violated up to a known bound. Such a violation may lead to non-satisfaction of the timeliness properties up to a certain bound. We characterise this bound in terms of (a) the longest interval when the deadlines are not met, which we call the settling-time, and (b) the maximum number of jobs that can miss deadlines during the settling-time called the overshoot. We propose two models of rare events, characterised on an interval domain. For a single stream of jobs, we provide methods to tightly compute the settling-time and overshoot. For multiple streams of jobs on a single processor, we show that amongst schedulers agnostic to the occurrence of the rare event, the EDF scheduler optimally minimises the settling-time. In contrast, RM is not optimal within the class of fixed priority schedulers.

An Algorithm for Online Reconfiguration of Resource Reservations for Hard Real-Time Systems

Nowadays, real-time applications expect the supporting computing system to be reconfigured at run-time. Even during such reconfiguration, timing requirements of the applications must be met. By extension, such requirements are relevant in the design of resource reservations techniques. In this work, we consider such a reconfiguration of the reservation provided by a constant bandwidth server (CBS). Firstly, we de-fine an exact notion of correctness of a servers reconfiguration. Then we design a provably correct server algorithm R-CBS that allows for run-time reconfiguration of a standard CBS. The algorithm maintains specific information about the execution trace and uses it to efficiently perform the reconfiguration at the earliest possible time. We highlight the advantages of R-CBS in comparison to reconfiguration of TDMA servers and in reconfiguring multiple servers simultaneously.

Demand bound server: generalized resource reservation for hard real-time systems

Servers have been proposed to implement resource reservations on shared resources. Such reservations isolate the temporal behavior of tasks sharing the shared resources, thereby providing performance guarantees to tasks independent of other tasks. In existing work, resource reservation has been synonymous to utilization (also called bandwidth) on the resource, i.e., we can reserve only a constant fraction of the resource utilization via a server. Such reservation schemes are not suited to serve interrupt-like tasks: tasks that occur seldom but require quick service or tasks with jitter. With this motivation, we present a generalized server algorithm, called Demand Bound Server (DBS), whose offered service is characterized by the demand bound function (dbf) of the task it serves. We show that schedulability of DBS tightly follows that of EDF, and if schedulable a DBS provides a performance guarantee as requested by the dbf of the task. We present an implementation of DBS when the dbf is a shifted-periodic curve and characterize its overhead. We also present efficient composition operations on DBS that widen the class of implemented servers to tightly serve tasks arising in most practical settings.