Papers by Hashem Ghazzawi
The emergence of complex real-time systems (RTS) can be widely witnessed in industry. We have org... more The emergence of complex real-time systems (RTS) can be widely witnessed in industry. We have organisational job handling across heterogeneous, largely independent and/or interdependent environments with high computing tasks to be shared across large areas/countries. We need techniques to assure scheduling tasks efficiently and meeting their deadlines within an acceptable range, overall stability, efficient monitoring and maintainability. Traditional real-time computing assumes worst-case requirements are known a priori. This will provide guarantees in avoidance of undesirable effects such as overload and deadline misses. In LSCITS, one of the main concerns is to design adaptation capabilities that handle uncertain effects dynamically and in an analytically predictable manner. Classical realtime scheduling algorithms such as EDF and RM are considered to be open loop techniques. Meaning that the end result is not fed back to the system to improve the performance (minimise the error). The benefit of feeding back and computing the system error is that we would agree on what actions to take to achieve the desired behaviour to acceptable range. Here, actions are either increasing the CPU/memory/network utilisations or assigning different priorities to tasks etc. In other words, these algorithms work appropriately in predicted environments where tasks execution times and deadlines are known a priori. This is not the case with LSCITS, the need of being able to monitor the system performance and constantly having knowledge of the errors margin is highly desired. Control theory offers feedback system modelling where it makes working in such an LSCITS environment is manageable and delivers acceptable results. Modelling a computing system as a dynamic system or as a controller is an approach that has proved to be fruitful in many cases.
In this paper, we propose a novel admission control scheduling (ACS) algorithm for scheduling sof... more In this paper, we propose a novel admission control scheduling (ACS) algorithm for scheduling soft real-time tasks with dependencies. Such tasks represent the workload of a particular industrial system where the real-time scheduling (RTS) objective is to ensure performance guarantees for maximum processor utilisation and slack values. Due to having dependencies between tasks, we now have the notion of jobs where a single job contains a number of dependent tasks. Thus, we propose looking into further RTS objectives and constraints due to tasks dependencies. We aim in this paper to provide a study on the efficiency of model predictive control-based ACS in handling different dependency shapes that can be inherent in industrial computer systems.
ABSTRACT This paper provides an empirical investigation between multiple control-theoretic approa... more ABSTRACT This paper provides an empirical investigation between multiple control-theoretic approaches in real-time scheduling systems. These approaches include proportional-integral-derivative (PID) and model predictive control. The former is a widely adopted controller due to its simplicity in design and implementation. The latter is considered a sophisticated one in the control community with respect to controller design. PID is widely adopted in industrial systems that are electrical and mechanical oriented. However, due to the advent of utilising control-theoretic approaches in real-time scheduling systems, sophisticated controllers have been argued to provide better performance with respect to meeting real-time multi-objective optimisation problems such as reducing deadline slacks and increasing CPU utilisation. We will also investigate multi-CPU platforms and cross view the performance of the different controllers.
ABSTRACT This paper explores the performance of feedback control when managing workflows in compu... more ABSTRACT This paper explores the performance of feedback control when managing workflows in computing systems. Industrial systems nowadays can consist of geographically diverse and heterogeneous high-performance computing (HPC) clusters. When scheduling workflows over such platforms, it is often desired to observe a number of real-time objectives such as meeting deadlines, reducing slacks, and increasing platform utilisation. We apply a control theoretic approach to address scheduling-related trade-offs of workflows that are executed in HPC platforms. Our results show that model predictive control-based admission controller is efficient for scheduling periodic workflows in a homogeneous HPC cluster with respect to minimum slacks and maximum CPU utilisation.
2014 9th International Symposium on Reconfigurable and Communication-Centric Systems-on-Chip (ReCoSoC), 2014
ABSTRACT This paper presents early exploration of the feedback admission control for high-perform... more ABSTRACT This paper presents early exploration of the feedback admission control for high-performance computing clusters executing real-time tasks using controlled values to dynamic task allocation. A number of controller variants with different architectures and relying on various metrics have been proposed. Simulation models of both open- and closed-loop systems have been prepared and compared. The obtained experimental results show that the proposed approach leads to executing almost five times more tasks before their deadlines in case of periodic uniform workload, and about 16% more for bursty workloads with large computation time variance.
Uploads
Papers by Hashem Ghazzawi