TL;DR
This paper provides a comprehensive survey of existing Network Function Virtualization (NFV) platforms, categorizing them based on their role in the network function lifecycle. It delves into the design choices made by these platforms, offering a valuable guide for operators and developers navigating the complex NFV landscape and anticipating future challenges in 5G and beyond.
What Problem Does This Paper Address?
Traditional network services rely on expensive, purpose-built hardware appliances that are inflexible and slow to update. Network Function Virtualization (NFV) aims to revolutionize this by replacing these hardware appliances with software-based Virtual Network Functions (VNFs) running on commodity hardware. This shift promises greater agility, cost-effectiveness, and faster innovation. However, the rapid development of numerous NFV platforms, each with its own unique design choices and motivations, has created a complex ecosystem. This makes it incredibly difficult for network operators to choose the right platform for their needs and for developers to build new ones. This paper addresses this challenge by providing a structured overview and analysis of the NFV platform design space.
Methods
The authors employ a systematic approach to survey and analyze NFV platforms. Their methodology involves:
- Architectural Overview: Presenting a generalized NFV platform architecture, combining the ETSI reference model with common industry implementations. This includes the MANO (Management and Orchestration) plane, the service plane, and the NFV Infrastructure (NFVI).
- Taxonomy Development: Classifying existing NFV platforms based on their primary focus within the typical lifecycle of a network function: prototyping, testing, deployment, and management/execution.
- Design Space Exploration: Thoroughly examining the various design choices made by platforms across different components (MANO plane, service plane, NFVI). This includes high-level APIs, execution models, packet I/O techniques, virtualization methods, and optimization strategies.
- Future Challenges Identification: Discussing emerging trends and potential hurdles for NFV platforms, particularly in the context of 5G, Artificial Intelligence (AI), network slicing, and the Internet of Things (IoT).
Datasets
Not discussed in the paper. The paper focuses on surveying existing platforms and their design choices rather than presenting new experimental results on specific datasets.
Key Insights and Ideas
- NFV Platform Diversity is a Double-Edged Sword: While the proliferation of NFV platforms offers choice and specialization, it also creates significant complexity for selection and implementation.
- Lifecycle-Centric Taxonomy is Crucial: Understanding where a platform excels within the VNF lifecycle (prototyping, testing, deployment, management) is key to choosing the right tool.
- Design Choices Matter Significantly: Decisions in areas like execution models (run-to-completion vs. pipeline), packet I/O (polling vs. interrupt), and virtualization techniques (VMs vs. containers) have a direct impact on performance, resource utilization, and security.
- The MANO Plane is Central: The Management and Orchestration (MANO) plane, encompassing the Orchestrator (NFVO), Virtualized Infrastructure Manager (VIM), and VNF Manager (VNFM), is the brain of an NFV system, dictating how services are provisioned and managed.
- Future Challenges are Significant: Integrating AI, managing network slices, and supporting the massive scale of IoT present complex challenges that NFV platforms will need to address.
Key Results
The paper's primary contribution is not a set of quantitative results from new experiments, but rather a comprehensive qualitative analysis and categorization of existing NFV platforms. Key findings include:
- A detailed taxonomy that categorizes numerous NFV platforms based on their focus within the VNF lifecycle.
- An extensive exploration of the design space, highlighting the trade-offs and common choices made by platforms in areas such as:
- MANO Plane: Use of GPL vs. DSL for APIs, various placement strategies, and different state coordination mechanisms.
- Service Plane: Preference for run-to-completion (RTC) execution models for simpler VNFs and pipeline models for finer-grained scaling, and the integration of TCP/IP stacks.
- NFVI: Widespread adoption of DPDK for high-performance packet I/O, and the use of VMs and containers as primary virtualization techniques.
- Identification of future challenges in AI integration, network slicing, and IoT support, emphasizing the need for new mechanisms to ensure security, scalability, and efficient resource management.
Limitations and Future Work
Limitations:
- The paper focuses on existing NFV platform implementations and does not introduce new experimental benchmarks or datasets.
- The rapidly evolving nature of NFV means that new platforms and design choices are constantly emerging, so the survey represents a snapshot in time.
Future Work:
- AI Integration: Further research is needed to effectively integrate AI techniques into NFV platforms for traffic prediction, runtime management, and automated optimization, while addressing the complexity and expertise required.
- Network Slicing: Developing robust mechanisms for dynamic slice instantiation/termination, cognitive resource scheduling, and adaptive VNF placement across slices is crucial for 5G.
- IoT Integration: NFV platforms need to be tailored for IoT use cases, addressing challenges related to massive device deployment, ultra-low latency, high throughput, privacy, and energy efficiency.
- Security: Enforcing security policies for slices spanning multiple administrative domains and ensuring isolation between slices are critical areas for future work.
Summary updated