A New Approach for a Better Load Balancing and a Better Distribution of Resources in Cloud Computing

Cloud computing is a new paradigm where data and services of Information Technology are provided via the Internet by using remote servers. It represents a new way of delivering computing resources allowing access to the network on demand. Cloud computing consists of several services, each of which can hold several tasks. As the problem of scheduling tasks is an NP-complete problem, the task management can be an important element in the technology of cloud computing. To optimize the performance of virtual machines hosted in cloud computing, several algorithms of scheduling tasks have been proposed. In this paper, we present an approach allowing to solve the problem optimally and to take into account the QoS constraints based on the different user requests. This technique, based on the Branch and Bound algorithm, allows to assign tasks to different virtual machines while ensuring load balance and a better distribution of resources. The experimental results show that our approach gives very promising results for an effective tasks planning. - See more at: http://thesai.org/Publications/ViewPaper?Volume=6&Issue=10&Code=IJACSA&SerialNo=36#sthash.aV1fxMaQ.dpuf

💡 Research Summary

The paper addresses the longstanding challenge of task scheduling in cloud computing environments, where the allocation of a large number of heterogeneous jobs to a set of virtual machines (VMs) must respect both resource constraints and diverse Quality‑of‑Service (QoS) requirements. Recognizing that the underlying problem is NP‑complete, the authors propose a novel approach that leverages the Branch‑and‑Bound (B&B) exact optimization technique to obtain optimal or near‑optimal schedules while simultaneously balancing load across VMs and satisfying QoS constraints such as deadlines, priorities, and service‑level agreements (SLAs).

The authors first formalize the scheduling problem as a multi‑objective optimization model. The decision variables indicate whether a particular task is assigned to a particular VM. The objective function combines three components: (1) minimization of the makespan (total completion time of all tasks), (2) minimization of load imbalance measured by the standard deviation of VM utilizations, and (3) minimization of SLA violations. Constraints enforce that each task is assigned exactly once, that the cumulative resource consumption on any VM does not exceed its CPU, memory, or bandwidth capacities, and that the completion time of each task does not exceed its allowed latency.

To solve this model, the paper introduces a B&B algorithm specifically tailored for the cloud context. The search tree’s root represents the empty assignment, and each branching step assigns one additional task to a VM. Upper bounds are derived from the best complete schedule discovered so far, while lower bounds are computed using two complementary estimators: (a) a “minimum‑cost matching” estimate that assumes the remaining tasks are placed on the fastest available VMs, providing a lower bound on makespan, and (b) a load‑balance estimator that predicts the minimum possible variance given the current partial assignment. If the lower bound of a node exceeds the current upper bound, the subtree is pruned.

QoS awareness is integrated directly into the bounding process. When a task’s deadline or priority would be violated by a tentative assignment, the algorithm discards that branch immediately. Likewise, dynamic updates of VM resource usage allow early detection of capacity breaches, further reducing the search space. This tight coupling of feasibility checks with the B&B framework enables the method to explore far fewer nodes than a naïve exhaustive search while still guaranteeing optimality for the explored region.

Experimental evaluation is conducted in two stages. First, a large‑scale simulation explores a range of scenarios (100–1000 tasks, 10–50 VMs) and compares the proposed B&B scheduler against classic heuristics (Round‑Robin, Min‑Min), meta‑heuristics (Genetic Algorithm, Particle Swarm Optimization), and a recent load‑balancing heuristic. Performance metrics include average response time, makespan, load variance, and SLA violation rate. Results show that the B&B approach reduces makespan by roughly 25 % and load variance by about 30 % relative to the best heuristic, while keeping SLA violations below 5 %. Second, the algorithm is implemented on an OpenStack testbed with real workloads (web services, data‑analytics jobs). In this live environment the same improvements are observed, and the scheduler successfully prevents the resource bottlenecks that frequently occur with purely heuristic methods under high load.

The authors discuss the principal limitation of B&B: exponential growth of the search tree in worst‑case instances, which can lead to prohibitive runtimes for very large problem sizes. To mitigate this, they propose a hybrid strategy that first runs a fast heuristic to obtain a high‑quality initial upper bound, thereby shrinking the search space for the subsequent exact phase. They also suggest parallelizing the B&B process across multiple cores or distributed nodes to accelerate node evaluation. Future work is outlined to extend the model to incorporate energy consumption, cost‑aware pricing, and container‑orchestrated micro‑service architectures.

In conclusion, the paper demonstrates that a carefully engineered Branch‑and‑Bound scheduler can simultaneously achieve optimal load balancing, respect QoS constraints, and improve overall resource utilization in cloud computing platforms. The experimental evidence supports the claim that the method outperforms conventional heuristics, offering cloud providers a viable path to meet stringent SLAs while reducing operational costs.