Optimal Joint Multiple Resource Allocation Method for Cloud Computing Environments

Cloud computing is a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources. To provide cloud computing services economically, it is important to optimize resource allocation under the assumption that the required resource can be taken from a shared resource pool. In addition, to be able to provide processing ability and storage capacity, it is necessary to allocate bandwidth to access them at the same time. This paper proposes an optimal resource allocation method for cloud computing environments. First, this paper develops a resource allocation model of cloud computing environments, assuming both processing ability and bandwidth are allocated simultaneously to each service request and rented out on an hourly basis. The allocated resources are dedicated to each service request. Next, this paper proposes an optimal joint multiple resource allocation method, based on the above resource allocation model. It is demonstrated by simulation evaluation that the proposed method can reduce the request loss probability and as a result, reduce the total resource required, compared with the conventional allocation method. Then, this paper defines basic principles and a measure for achieving fair resource allocation among multiple users in a cloud computing environment, and proposes a fair joint multiple resource allocation method. It is demonstrated by simulation evaluations that the proposed method enables the fair resource allocation among multiple users without a large decline in resource efficiency. Keywords: Cloud computing, joint multiple resource allocation, fairness

💡 Research Summary

The paper addresses a fundamental challenge in cloud computing: allocating multiple resource types—specifically processing power (CPU) and network bandwidth—simultaneously to incoming service requests while minimizing waste and ensuring fairness among users. The authors first construct a formal resource‑allocation model for a distributed cloud environment consisting of k geographically separated data centers. Each center j possesses a maximum processing capacity Cmax j and a maximum bandwidth Nmax j. Service requests arrive with a pair of demands (C, N) representing the required CPU share and bandwidth, respectively, and are assumed to be non‑delay (i.e., they must be served immediately or be rejected). Once allocated, the resources are dedicated to the request for a fixed service time H, after which both CPU and bandwidth are released together.

Two allocation strategies are compared. Method I follows a simple round‑robin policy: centers are examined in a pre‑defined order, and the first one with sufficient CPU and bandwidth is chosen; if none can satisfy the request, it is dropped. This approach ignores the interdependence of the two resource dimensions and can lead to a “deadlock” situation where enough total resources exist in the system, but no single center has both CPU and bandwidth simultaneously available, causing unnecessary request loss.

To overcome this limitation, the authors propose Method II, a “identified‑resource” based optimal joint allocation algorithm. For each request, the required CPU and bandwidth are normalized by the smallest maximum among all centers, yielding ratios XC = C_req / C0 and XN = N_req / N0. The resource with the larger ratio is designated the identified resource (the one that is most “constrained” for that request). The algorithm then selects the center whose remaining amount of the identified resource is the smallest among those that can still satisfy both CPU and bandwidth demands (a best‑fit heuristic). If multiple centers meet this criterion, one is chosen at random. The selected center receives both resources simultaneously, and they are released together after the service time expires.

Simulation experiments are conducted using a C‑language discrete‑event simulator. The test scenario involves two centers (k = 2) with Gaussian‑distributed CPU and bandwidth demands and exponentially distributed inter‑arrival times. Results show that Method II dramatically reduces the request loss probability compared with Method I and achieves the same level of service with roughly 10–20 % fewer total resources, confirming that jointly considering both dimensions prevents the deadlock effect and improves overall utilization.

Beyond efficiency, the paper tackles fairness. When multiple tenants compete for resources, a naïve optimal algorithm may favor those with larger or more frequent requests, leading to inequitable allocations. The authors define a fairness metric based on the normalized cumulative allocation to each user and aim to minimize the variance of this metric across users. They extend Method II with a fairness‑aware scheduling component, termed Fair Joint Allocation, which adjusts the selection order to balance the fairness metric while still respecting the identified‑resource best‑fit rule. Simulations indicate that this fairness‑enhanced version substantially lowers the disparity among users with only a modest (~5 %) reduction in overall resource efficiency.

In summary, the contributions of the paper are threefold: (1) a rigorous multi‑resource allocation model for distributed cloud environments; (2) an identified‑resource best‑fit algorithm (Method II) that jointly allocates CPU and bandwidth, eliminating deadlock‑induced request loss and reducing required capacity; and (3) a fairness‑oriented extension that equalizes user allocations without sacrificing much efficiency. The work provides a practical pathway for cloud providers to cut operational costs while delivering more reliable and equitable service. Future research directions suggested include incorporating dynamic migration, variable service durations, and additional resource types such as storage or GPUs to further generalize the framework.