In the current era of distributed computing, the "cloud" is not a single, unified utility but a collection of isolated islands. While users can rent virtually unlimited compute power, they are often trapped within the proprietary ecosystems of a few dominant providers. This phenomenon, known as vendor lock-in, is the central problem that a growing body of research seeks to solve through the development of cross-cloud abstraction layers.
The goal of this research direction is to move from "Cloud Computing"—where applications are tethered to specific provider APIs—to "Sky Computing," a vision of a truly interoperable global computing utility.
The Gravity of Proprietary Silos
The primary barrier to a unified cloud is what Yehia Elkhatib describes as "operational gravity" in Mapping Cross-Cloud Systems: Challenges and Opportunities. As developers integrate high-level proprietary services—such as managed databases, specialized AI pipelines, or unique serverless orchestration—the cost and complexity of migrating to another provider grow exponentially.
This gravity is reinforced by a lack of standardization. As Fei Hu and colleagues note in A Review on Cloud Computing: Design Challenges in Architecture and Security, the absence of unified industry standards for cloud architecture prevents users from effectively migrating workloads. This isn't just a technical oversight; Yehia Elkhatib argues in Defining Cross-Cloud Systems that major providers have a commercial incentive to resist universal standards to protect their market share and prevent customer churn.
Decoupling Logic from Infrastructure
To break this lock-in, researchers are developing mechanisms that decouple an application’s logic from the underlying cloud infrastructure. This is achieved through three primary layers of abstraction:
- Containerization as the "Glue": Muhammad Waseem and co-authors identify containers as the fundamental unit of portability in Containerization in Multi-Cloud Environment. By packaging code and dependencies together, containers provide a consistent environment across different clouds.
- Declarative Orchestration: Tools like TORCH, developed by Orazio Tomarchio and Giuseppe Modica, use the TOSCA (Topology and Orchestration Specification for Cloud Applications) standard to describe application deployments in a provider-agnostic way. This allows a "design once, deploy anywhere" workflow, as detailed in TORCH: a TOSCA-Based Orchestrator of Multi-Cloud Containerised Applications.
- Unified Service Metadata: For complex analytics, Gregor von Laszewski proposes the "AS-FAIR-DO" standard in Whitepaper on Reusable Hybrid and Multi-Cloud Analytics Service Framework. This framework ensures that services are not just portable, but findable and operational across heterogeneous infrastructures.
From Cloud to Sky Computing
The culmination of these efforts is the concept of Sky Computing. In the white paper The Sky Above The Clouds, Alvin Cheung, Ali Ghodsi 0002, and others at UC Berkeley envision a "compatibility layer" that sits above individual clouds.
This layer acts similarly to the "narrow waist" of the Internet (the IP protocol). Just as IP allows different physical networks to communicate, a Sky Computing layer would allow applications to be dispatched to any cloud provider based on cost, performance, or policy, without the developer needing to know the specific APIs of AWS, Azure, or GCP.
Why This Matters for Science and Industry
The push for interoperability is particularly urgent in large-scale scientific research. Ian T. Foster and Péter Kacsuk highlight in their Editors' Message that achieving provider independence is essential for integrating proprietary cloud services with local high-performance computing (HPC) resources.
In fields like High Energy Physics, the lack of standardization creates significant friction. As noted in Analysis Facilities for the HL-LHC White Paper, researchers struggle to move analysis tasks between disparate cloud-native and legacy environments because of varying orchestration layers and storage protocols.
By building these abstraction layers, the research community is attempting to transform the cloud from a fragmented market of proprietary services into a seamless, interoperable utility that can support the next generation of global-scale computation.