Hafiz Fahad Sheikh

Energy- and performance-aware scheduling of tasks on parallel and distributed systems

Enabled by high-speed networking in commercial, scientific, and government settings, the realm of high performance is burgeoning with greater amounts of computational and storage resources. Large-scale systems such as computational grids consume a significant amount of energy due to their massive sizes. The energy and cooling costs of such systems are often comparable to the procurement costs over a year period. In this survey, we will discuss allocation and scheduling algorithms, systems, and software for reducing power and energy dissipation of workflows on the target platforms of single processors, multicore processors, and distributed systems. Furthermore, recent research achievements will be investigated that deal with power and energy efficiency via different power management techniques and application scheduling algorithms. The article provides a comprehensive presentation of the architectural, software, and algorithmic issues for energy-aware scheduling of workflows on single, multicore, and parallel architectures. It also includes a systematic taxonomy of the algorithms developed in the literature based on the overall optimization goals and characteristics of applications.

Stretch and compress based re-scheduling techniques for minimizing the execution times of DAGs on multi-core processors under energy constraints

Given an initial schedule of a parallel program represented by a directed acyclic graph (DAG) and an energy constraint, the question arises how to effectively determine what nodes (tasks) can be penalized (slowed down) through the use of dynamic voltage scaling. The resulting re-schedule length with a strict energy budget should have a minimum amount of expansion compared to the original schedule achieved with full energy. We propose three static schemes that aim to achieve this goal. Each scheme encompasses submitting a schedule to either a conceptual “stretch” (starting tasks with a maximum voltage supplied to all cores followed by methodical voltage reductions) or “compress” (starting tasks with a minimum voltage supplied to all cores followed by methodical voltage boosts). The complexity arises due to the inter-dependence of tasks. We propose methods that efficiently make such findings by analyzing the DAG and determining the “impact factor” of a node in the graph for the purpose of guiding the schedule toward the desired goal. The comparison between the stretch-alone and compress-alone based algorithms leads to a third algorithm that employs schedule “compression,” but reschedules all cores following each successive voltage adjustment. Detailed simulation experiments demonstrate the effect of various task and processor parameters on the performance of the proposed algorithms.

Simultaneous optimization of performance, energy and temperature for DAG scheduling in multi-core processors

This paper addresses the joint optimization of performance, energy, and temperature, termed as PET - optimization. This multi-objective PET-optimization is achieved in scheduling DAGs on multi-core systems. Our technique is based on multi-objective evolutionary algorithm (MOEA) for finding Pareto optimal solutions using scheduling and voltage selection. These solutions are not necessarily scalar values but can be in a vector form. We developed a Strength Pareto Evolutionary Algorithm [2] (SPEA) based solution which is inherently superior to several other MOEA methods. The proposed algorithm obtains the Pareto vectors (or fronts) efficiently. The work is novel and original in the sense that no previous such optimization work has been reported to our knowledge for the PET-optimization scheduling problem. The strength of the proposed algorithm is that it achieves diverse range of energy and thermal improvements while staying close to the performance-optimal point to ensure efficient trade-off solutions. The proposed approach consists of two-steps. In the first step, Pareto fronts are generated. In the second step, one most optimal solution is selected. Simulation results on several benchmark task graph applications demonstrate that efficient solutions can be selected using the proposed selection method in polynomial time.

An Evolutionary Technique for Performance-Energy-Temperature Optimized Scheduling of Parallel Tasks on Multi-Core Processors

This paper proposes a multi-objective evolutionary algorithm (MOEA)-based task scheduling approach for determining Pareto optimal solutions with simultaneous optimization of performance (P), energy (E), and temperature (T). Our algorithm includes problem-specific solution encoding, determining the initial population of the solution space, and the genetic operators that collectively work on generating efficient solutions in fast turnaround time. Multiple schedules offer a diverse range of values for makespan, energy consumed, and peak temperature and thus present an efficient way of identifying trade-offs among the desired objectives, for a given application and machine pair. We also present a methodology for selecting one solution from the Pareto front given the users preference. The proposed algorithm for scheduling tasks to cores achieves three-way optimization with fast turnaround time. The proposed algorithm is advantageous because it reduces both energy and temperature together rather than in isolation. We evaluate the proposed algorithm using implementation and simulation, and compare it with integer linear programming as well as with other scheduling algorithms that are energyor thermal-aware. The time complexity of the proposed scheme is considerably better than the compared algorithms.

Efficient heuristics for joint optimization of performance, energy, and temperature in allocating tasks to multi-core processors

This paper presents heuristic algorithms for solving the three-way joint optimization of Performance, Energy and temperature (PET) in scheduling tasks to multi-core processors. The problem, called as PET optimized scheduling (PETOS) problem is a high-complexity problem due to conflicting objectives. While solutions to the PETOS problem can be obtained by using conventional multi-objective optimization approaches, the time taken by such solvers is generally not feasible to be used at the scheduling level. Therefore, we explore heuristic methods that can explore the decision space while maintaining low computational complexity. The design of heuristic algorithms presents non-trivial challenges of incorporating all the PET quantities into the scheduling process. We present nine heuristics, each varying in its approach for selecting a core and picking the processor frequency. Each heuristic produces a set of solutions where each solution represents a complete schedule for assigning a set of tasks on a given multi-core system, thus identifying different trade-offs that exist between performance, energy, and temperature at scheduling level. A comparative study describes their trade-offs.

Dynamic task graph scheduling on multicore processors for performance, energy, and temperature optimization

Despite significant advancements in multicore processor technology for reducing the chip-level energy consumption, higher levels of power dissipation resulting in thermal implications and cooling costs remain as unsolved problems. Although several scheduling methods of controlling and managing the power dissipation and temperature exist, most schemes are static that are unable to adjust to the dynamic program and system changes. This paper presents dynamic method for voltage-scaling based task scheduling for simultaneous optimization of performance, energy, and temperature (PET quantities) under dynamically varying task and system conditions. Our method generates an initial set of Pareto optimal solutions utilizing a multi-objective evolutionary algorithm (MOEA) called SPEA-II (Strength Pareto Evolutionary Algorithm). This set of solutions is dynamically evolved with time to minimize the deviation of PET quantities from the Pareto optimal values. We carried out extensive evaluations using several task graph benchmarks based on the data obtained from a real multicore machine. The results indicate that the proposed dynamic re-optimization achieves up to 8% improvement in PET quantities as compared to the statically selected schedule.

Fast algorithms for thermal constrained performance optimization in DAG scheduling on multi-core processors

Thermal management is highly crucial for efficient exploitation of the potentially enormous computational power offered by advanced multi-core processors. Higher temperatures can adversely affect these processors. Without any thermal constraint, a task graph may be scheduled to run on the cores at their maximum voltage. Very often, multiple factors lead to imposing constraints on temperature, ensuring that cores remain below a certain temperature range and yet deliver good performance. The challenge is how to schedule the same task graph under the imposed thermal constraints such that the performance degradation is the minimum. In this paper we present two algorithms for minimizing the performance degradation and the corresponding overhead while satisfying the thermal constraints. The proposed algorithms, named PAVD, and TAVD, adjust a given schedule of a task graph by decreasing the voltage level of judiciously selected tasks in each step. The algorithms differ in the way they select a task at each step and the amount of time spent in searching that task. TAVD selects the tasks by prioritizing among the cores and tasks which attained maximum temperature while PAVD selects the tasks with the minimum performance penalty. For comparison, we develop a simpler greedy-based approach to show that the problem is non-trivial. Extensive experiments using both random and application-oriented task graphs demonstrate that all three algorithms satisfy the imposed thermal constraints by trading-off performance, while each showing its own strength.

Optimizing performance and energy in computational grids using non-cooperative game theory

There is a lack of generally applicable methods for reducing energy consumption while ensuring good quality of service in distributed computational grids. We study the energy-aware task allocation problem for assigning a set of tasks onto the machines in a grid environment where the conflicting goals of ensuring quality of service and reducing energy consumption makes the machines compete with each other. We propose bidding mechanisms in which the machines have to win in order to maintain a minimum fitness value and thus remain relevant to the system and hence must try their best to meet the goals. The grid manager keeps only those machines that win and eliminate from the pool the ones that are unfit. The proposed algorithm encompasses bidding strategies, fitness calculations, penalties, exit as well as resurrection mechanisms to support a non-cooperative game in which all machines compete to win tasks. The concept of fitness is fundamental to our algorithm, defining a machines ability to remain in the system. When heterogeneous machines are part of a shared computing resource pool governed by a grid economy, the proposed approach fits very well for achieving conflicting goals. By simulating several machines with diverse architectures and task sets with varying requirements, we demonstrate the effectiveness of the proposed scheme and show that it generates short task makespans and reduced energy consumption. The algorithm is extremely fast, takes highly detailed machine and task characteristics into consideration, and outperforms the Earliest Deadline First Scheme in every aspect.

Sixteen Heuristics for Joint Optimization of Performance, Energy, and Temperature in Allocating Tasks to Multi-Cores

Three-way joint optimization of performance (P), energy (E), and temperature (T) in scheduling parallel tasks to multiple cores poses a challenge that is staggering in its computational complexity. The goal of the PET optimized scheduling (PETOS) problem is to minimize three quantities: the completion time of a task graph, the total energy consumption, and the peak temperature of the system. Algorithms based on conventional multi-objective optimization techniques can be designed for solving the PETOS problem. But their execution times are exceedingly high and hence their applicability is restricted merely to problems of modest size. Exacerbating the problem is the solution space that is typically a Pareto front since no single solution can be strictly best along all three objectives. Thus, not only is the absolute quality of the solutions important but “the spread of the solutions” along each objective and the distribution of solutions within the generated tradeoff front are also desired. A natural alternative is to design efficient heuristic algorithms that can generate good solutions as well as good spreads -- note that most of the prior work in energy-efficient task allocation is predominantly single- or dual-objective oriented. Given a directed acyclic graph (DAG) representing a parallel program, a heuristic encompasses policies as to what tasks should go to what cores and at what frequency should that core operate. Various policies, such as greedy, iterative, and probabilistic, can be employed. However, the choice and usage of these policies can influence a heuristic towards a particular objective and can also profoundly impact its performance. This article proposes 16 heuristics that utilize various methods for task-to-core allocation and frequency selection. This article also presents a methodical classification scheme which not only categorizes the proposed heuristics but can also accommodate additional heuristics. Extensive simulation experiments compare these algorithms while shedding light on their strengths and tradeoffs.

A comparison of evolutionary techniques for task-to-core scheduling algorithms with performance, energy, and temperature optimization

Performance, energy and temperature (PET) are closely related and must be considered holistically while addressing the balance between them. Multi-objective evolutionary algorithms (MOEAs) for finding Pareto optimal solutions are highly effective in generating solutions for task-to-core scheduling and voltage selection on individual cores. A solution set comprises of multiple points forming a vector or a front, not just scalar values. Evolutionary techniques such as Strength Pareto Evolutionary Algorithm (SPEA) and Non-Dominated Sorting Genetic Algorithm (NSGA), are superior optimization methods used in many scenarios. This paper utilizes these techniques in PET-based scheduling algorithms, describes an NSGA-based algorithm, and highlights the difference between the two approaches. By using a set of benchmarks, evaluation procedures, performance measures, the paper compares the two scheduling algorithms in evaluating trade-offs and determining which parameters affect the results. Extensive experimentation carried out in addressing the above issues facilitated the comparison of the two algorithms amongst themselves as well as with optimal solutions obtained through Integer Linear Programming.