[Paper Review] Managing Deadline-constrained Bag-of-Tasks Jobs on Hybrid Clouds

Bo Wang, Ying Song+2 more|arXiv (Cornell University)|Apr 28, 2016

Cloud Computing and Resource Management16 references9 citations

TL;DR

This paper proposes a heuristic algorithm for scheduling deadline-constrained bag-of-tasks (BoT) jobs on hybrid clouds to minimize public cloud rent costs while maximizing local resource utilization. By prioritizing tasks closest to their deadline and dynamically provisioning the most cost-effective physical machines or VMs, the method reduces rent costs by 16.2%–76% and improves resource utilization by 47.3%–182.8% compared to first-fit decreasing.

ABSTRACT

Outsourcing jobs to a public cloud is a cost-effective way to address the problem of satisfying the peak resource demand when the local cloud has insufficient resources. In this paper, we study on managing deadline-constrained bag-of-tasks jobs on hybrid clouds. We present a binary nonlinear programming (BNP) problem to model the hybrid cloud management where the utilization of physical machines (PMs) in the local cloud/cluster is maximized when the local resources are enough to satisfy the deadline constraints of jobs, while when not, the rent cost from the public cloud is minimized. To solve this BNP problem in polynomial time, we proposed a heuristic algorithm. Its main idea is assigning the task closest to its deadline to current core until the core cannot finish any task within its deadline. When there is no available core, the algorithm adds an available PM with most capacity or rents a new VM with highest cost-performance ratio. Extensive experimental results show that our heuristic algorithm saves 16.2%-76% rent cost and improves 47.3%-182.8% resource utilizations satisfying deadline constraints, compared with first fit decreasing algorithm.

Motivation & Objective

• To address the challenge of cost-efficiently managing deadline-constrained bag-of-tasks (BoT) jobs in hybrid clouds with heterogeneous local resources.
• To minimize the cost of leasing virtual machines (VMs) from public clouds while maximizing utilization of local physical machines (PMs).
• To develop a polynomial-time heuristic that effectively maps tasks to resources under strict deadline constraints.
• To evaluate the algorithm’s performance against the first-fit decreasing (FFD) baseline using real-world workloads.
• To balance cost minimization and resource utilization in hybrid cloud environments for scientific and data-intensive workloads.

Proposed method

• Formulates the hybrid cloud management problem as a binary nonlinear programming (BNP) model to jointly optimize local PM utilization and public cloud rent cost.
• Proposes a heuristic algorithm that assigns tasks to cores based on minimizing the difference between finish time and deadline.
• When no task can be completed on existing resources within deadline, the algorithm adds the PM with the highest available capacity or leases the VM with the best cost-performance ratio.
• The algorithm dynamically expands the resource pool by selecting the most efficient local PM or public cloud VM at each step.
• Uses a two-phase strategy: first, assign tasks to existing cores to minimize deadline deviation; second, provision new resources only when necessary.
• Employs a cost-performance ratio metric to prioritize VM leasing, ensuring cost efficiency in public cloud extensions.

Experimental results

Research questions

• RQ1 How can task scheduling and resource provisioning be jointly optimized in hybrid clouds to minimize public cloud leasing costs for deadline-constrained BoT jobs?
• RQ2 What is the impact of heterogeneous local resources on the efficiency and cost of hybrid cloud job management?
• RQ3 Can a heuristic algorithm achieve near-optimal performance in polynomial time for the NP-hard BNP formulation of hybrid cloud management?
• RQ4 How does the proposed algorithm compare to the first-fit decreasing (FFD) algorithm in terms of cost savings and resource utilization?
• RQ5 To what extent can local resource utilization be maximized while ensuring all jobs meet their deadlines?

Key findings

• The proposed heuristic algorithm reduces rent costs by 16.2% to 76% compared to the first-fit decreasing (FFD) algorithm across two real work traces.
• Resource utilization under deadline constraints improves by 47.3% to 182.8% compared to FFD, indicating superior efficiency in local resource usage.
• The algorithm achieves high scalability and performs well under diverse workload patterns and deadline constraints.
• The BNP model effectively captures the trade-off between local resource utilization and public cloud cost, enabling cost-efficient hybrid cloud management.
• The heuristic solution is computationally efficient and runs in polynomial time, making it practical for real-world deployment.
• The algorithm outperforms FFD by prioritizing tasks closest to their deadline and intelligently provisioning the most cost-effective resources.