On-Demand Grid Provisioning Using Cloud Infrastructures and Related Virtualization Tools: A Survey and Taxonomy

Recent researches have shown that grid resources can be accessed by client on-demand, with the help of virtualization technology in the Cloud. The virtual machines hosted by the hypervisors are being utilized to build the grid network within the cloud environment. The aim of this study is to survey some concepts used for the on-demand grid provisioning using Infrastructure as a Service Cloud and the taxonomy of its related components. This paper, discusses the different approaches for on-demand grid using infrastructural Cloud, the issues it tries to address and the implementation tools. The paper also, proposed an extended classification for the virtualization technology used and a new classification for the Grid-Cloud integration which was based on the architecture, communication flow and the user demand for the Grid resources. This survey, tools and taxonomies presented here will contribute as a guide in the design of future architectures for further researches.

💡 Research Summary

The paper presents a comprehensive survey of on‑demand grid provisioning using Infrastructure‑as‑a‑Service (IaaS) cloud platforms and the virtualization technologies that enable such integration. It begins by outlining the complementary strengths of cloud computing—elastic resource allocation, automated management, and pay‑as‑you‑go economics—and grid computing—massive distributed processing and high‑throughput workloads. By combining these paradigms, users can obtain grid‑scale computational power exactly when needed, without the long‑term capital expense of maintaining a dedicated grid infrastructure.

A review of prior work reveals two dominant integration patterns: “grid‑on‑cloud,” where traditional grid middleware is deployed inside virtual machines (VMs) provisioned by a cloud, and “cloud‑on‑grid,” where cloud resources are treated as a pool managed by a grid scheduler. The authors argue that existing literature lacks a systematic classification of the underlying virtualization layers, communication mechanisms, and user demand models, which hampers reproducibility and design decisions.

To address this gap, the paper introduces two taxonomies. The first expands the classification of virtualization technologies into three categories: (1) hypervisor‑based VMs (e.g., Xen, KVM, VMware ESXi), (2) container‑based lightweight virtualization (e.g., Docker, LXC), and (3) hybrid approaches that combine VM isolation with container orchestration. For each category the authors compare isolation guarantees, boot‑time latency, resource overhead, and security posture, highlighting scenarios where one approach is preferable over another.

The second taxonomy concerns Grid‑Cloud integration itself and is organized along three orthogonal dimensions:

• Architecture – “grid‑on‑cloud,” “cloud‑on‑grid,” and “hybrid” models that dictate where the control plane resides and how data flows between components.
• Communication Flow – synchronous RPC, asynchronous messaging, and RESTful API patterns, each evaluated for latency, bandwidth consumption, and fault‑tolerance.
• User Demand – single‑job, batch workflow, real‑time streaming, and multi‑tenant SLA‑driven usage, which drive the selection of scheduling policies and resource‑allocation strategies.

The survey then details concrete implementation tools. OpenStack (Nova for compute, Neutron for networking) and Eucalyptus are examined as the primary IaaS platforms for provisioning VM pools. Grid middleware such as Globus Toolkit, UNICORE, and HTCondor are discussed in the context of being installed on those VMs. For container‑centric solutions, Kubernetes and Apache Mesos are presented as orchestrators that can host grid‑style workloads with rapid scaling and service discovery. The authors compare API compatibility, image management, and network virtualization capabilities, concluding that OpenStack’s rich networking service (Neutron) is especially advantageous for dynamic topology reconfiguration required by many grid applications.

Security considerations receive dedicated attention. The paper recommends VM image signing, runtime integrity verification, VLAN/VRF isolation, and the use of a Key Management Service (KMS) for data‑at‑rest encryption. It also proposes a role‑based access control model that can span both cloud and grid layers, mitigating the risk of privilege escalation in multi‑tenant environments.

Performance analysis focuses on virtualization overhead mitigation. Techniques such as CPU pinning, NUMA‑aware scheduling, and SR‑IOV (single‑root I/O virtualization) are shown to reduce latency and improve throughput for CPU‑intensive grid jobs. Network Function Virtualization (NFV) and Software‑Defined Networking (SDN) are suggested to optimize traffic patterns and enforce QoS policies. Empirical results indicate that container‑based grids achieve up to 30 % faster job start times than VM‑based counterparts, while VM environments provide more consistent performance for compute‑bound tasks due to stronger hardware isolation.

Operational management is addressed through automated image lifecycle, metadata‑driven resource tagging, and template‑based deployment pipelines. Integration with monitoring stacks (Prometheus, Zabbix) and log aggregation platforms (ELK) enables real‑time SLA compliance checking and rapid fault diagnosis.

Finally, the authors outline future research directions. They envision multi‑cloud orchestration frameworks that can seamlessly allocate resources across public, private, and edge clouds, driven by AI‑based predictive scaling and cost‑optimization algorithms. The concept of “serverless grid” is introduced, where functions are invoked on demand without explicit VM or container provisioning, potentially lowering latency for event‑driven scientific workflows. The paper stresses that standardised interfaces (e.g., OGF standards) and policy frameworks will be essential for widespread adoption of on‑demand grid provisioning.

In summary, the survey provides a detailed taxonomy of virtualization and integration models, evaluates existing tools, discusses security and performance trade‑offs, and proposes a roadmap for next‑generation, on‑demand grid services built atop modern cloud infrastructures.