580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can online and hierarchical scheduling models effectively allocate jobs across multiple machines in Grid environments to minimize makespan?
This research area focuses on the design, analysis, and implementation of online scheduling algorithms tailored for Grids composed of multiple machines (clusters) with identical processors within each machine. It investigates non-preemptive scheduling of parallel jobs submitted over time, emphasizing allocation strategies that consider the structural constraints of Grids — notably that jobs run exclusively on processors within a single machine. The goal is to minimize makespan while managing the dynamic arrival of jobs and heterogeneity across machines. Hierarchical models separate scheduling into job-to-machine allocation followed by local scheduling, addressing practical limitations such as overheads in job migration and decentralized control. These models matter because efficient scheduling directly impacts Grid resource utilization, user satisfaction, and system throughput in highly distributed computational environments.
Key finding: Introduces a model extending classic multiprocessor scheduling by structuring processors into multiple machines, where jobs must execute wholly within one machine. Demonstrates that Garey and Graham’s list scheduling performs... Read more
Key finding: Proposes a two-stage online scheduling model where jobs are first assigned to machines and then locally scheduled. Introduces an adaptive admissible allocation algorithm and provides competitive analysis showing improved load... Read more
Key finding: Applies particle swarm optimization (PSO) meta-heuristics to centralized scheduling of independent jobs on heterogeneous Grid resources, noting that PSO tends to stagnate in local minima for high-dimensional problems. The... Read more
Key finding: Develops a novel hybrid meta-heuristic combining PSO and GA for centralized scheduling of independent jobs on a static Grid model, addressing PSO’s tendency to stagnate locally in large problem instances. Experimental... Read more
2. What is the comparative performance and theoretical basis of immediate (online) versus batch scheduling heuristics in Grid resource allocation?
This theme concerns the evaluation and analysis of heuristic scheduling approaches that decide how and when tasks/jobs are assigned to computational resources in Grid environments. Immediate (online) scheduling algorithms allocate tasks as soon as they arrive, which suits dynamic, real-time scenarios but may yield suboptimal load balancing. Batch scheduling collects a set of tasks and schedules them jointly at specific intervals, usually resulting in better makespan and resource utilization but incurring scheduling delay. Understanding the applicability, strengths, weaknesses, and trade-offs among heuristics such as Minimum Execution Time (MET), Minimum Completion Time (MCT), Min-Min, Max-Min, and their variants informs the design of practical Grid schedulers that balance efficiency, responsiveness, and load balance.
Key finding: Provides a detailed review and comparative analysis of immediate (online) scheduling heuristics like MET and MCT, alongside batch mode heuristics like Min-Min and Max-Min, highlighting their algorithmic principles and... Read more
Key finding: Proposes Max-Average, an enhanced Max-Min scheduling algorithm that improves load balancing and makespan minimization by combining characteristics of Max-Min and Min-Min heuristics. The algorithm operates in two phases:... Read more
Performance comparison of least slack time based heuristics for job scheduling on computational grid
Key finding: Develops and empirically evaluates hybrid scheduling algorithms combining deadline and slack time parameters in soft real-time Grid environments. By testing against real workload traces, the study finds that integrating slack... Read more
3. How do groupings and structural characteristics of job submissions affect scheduling performance and resource utilization in Grid systems?
This research domain examines the patterns in which jobs are submitted to Grid systems—such as batch submissions, continued sequences, or bursty job arrivals—and their implications on scheduling efficiency, resource consumption, and user experience. Characterizing and understanding such grouped job behaviors are essential to optimize scheduling policies, load balancing, and performance prediction. The focus lies on revealing real workload properties and how such groupings affect takeover scheduling mechanisms, slowdown metrics, and effective CPU utilization, contributing practical insights for improving Grid workload management.
Key finding: Analyzes three types of job groupings—batch, continued, and bursty submissions—in traces from production and experimental Grids, confirming that grouped jobs account for up to 96% of total CPU consumption. The study reveals... Read more