PaperlayerView exact revision on arXiv

Source-linked AI summary

Software-Defined Networking: State of the Art and Research Challenges

Manar Jammal, Taranpreet Singh, Abdallah Shami, Rasool Asal, Yiming Li

arXiv:1406.0124v1cs.NI

TL;DR

Legacy networks struggle with rapidly changing traffic, services, and infrastructure demands, motivating more flexible network management. The paper surveys SDN architecture, applications, benefits, challenges, and proposed solutions, concluding that SDN offers significant opportunities while reliability, scalability, latency, and security remain obstacles.

  • Problem

    Legacy networks face increasing traffic and management demands while their rigid, device-level control limits network-wide flexibility and efficiency.

  • Method

    The paper surveys SDN architecture, OpenFlow, network virtualization, applications in data centers and NaaS, technical challenges, and existing solutions.

  • Results

    The paper describes SDN benefits including programmable network control, simplified management, virtualization, and applications across data centers and NaaS.

  • Takeaways & Limitations

    SDN offers opportunities for more programmable and agile networks, but realizing them requires addressing reliability, scalability, latency, and security challenges.

  • Takeaways & Limitations

    SDN adoption remains constrained by reliability, scalability, latency, and security challenges.

Abstract

from arXiv · show

Plug-and-play information technology (IT) infrastructure has been expanding very rapidly in recent years. With the advent of cloud computing, many ecosystem and business paradigms are encountering potential changes and may be able to eliminate their IT infrastructure maintenance processes. Real-time performance and high availability requirements have induced telecom networks to adopt the new concepts of the cloud model: software-defined networking (SDN) and network function virtualization (NFV). NFV introduces and deploys new network functions in an open and standardized IT environment, while SDN aims to transform the way networks function. SDN and NFV are complementary technologies; they do not depend on each other. However, both concepts can be merged and have the potential to mitigate the challenges of legacy networks. In this paper, our aim is to describe the benefits of using SDN in a multitude of environments such as in data centers, data center networks, and Network as Service offerings. We also present the various challenges facing SDN, from scalability to reliability and security concerns, and discuss existing solutions to these challenges.

1. INTRODUCTION

Rapidly growing traffic, virtualization, and service demands expose the rigidity and management complexity of legacy networks. The paper presents SDN as a programmable, centralized-control approach and surveys its benefits and technical challenges.

  • 1. INTRODUCTION: Legacy networks collocate forwarding and decision-making, distribute intelligence across devices, and provide limited network-wide control.Their rigid, device-specific structure complicates management and restricts efficient adaptation to varied client requirements.
  • 1. INTRODUCTION: Video, big data, mobile usage, servers, and virtual machines are rapidly increasing traffic and operational demands for network operators.Operators require networks that are efficient and flexible to address these changes.
  • 1. INTRODUCTION: SDN separates the forwarding/data plane from a programmable, aggregated control plane to simplify network management and control.This architecture provides centralized visibility and enables software-based network changes.
  • 1. INTRODUCTION: The paper surveys SDN benefits and comprehensively discusses reliability, scalability, latency, controller placement, hardware shortages, and security challenges.It also reviews existing solutions and research initiatives across SDN environments and applications.

2. SOFTWARE-DEFINED NETWORKING AND OPENFLOW ARCHITECTURE

SDN separates network control from forwarding hardware, using controllers, APIs, and southbound protocols such as OpenFlow to enable programmable, network-wide management. Its benefits include simplified administration, virtualization, experimentation, and flexible control, while scalability and other technical challenges remain.

  • 2.1 SDN Architecture: SDN separates control and data-flow functionality, giving administrators an abstract network view for applying policies across hardware.Its architecture adds a controller, application interfaces, inter-controller protocols, and data-plane southbound protocols.
  • 2.2 SDN Benefits: SDN is intended to address legacy-network challenges through programmability, virtualization, centralized configuration, and improved visibility.The paper identifies reliability, scalability, latency, and security as continuing obstacles to adoption.
  • 2.2 SDN Benefits: SDN programmability abstracts flow control from individual devices to the network level, simplifying management and enabling network-wide policies.An external controller disseminates changes to selected devices or across network hardware over a secure channel.
  • 2.2 SDN Benefits: Network virtualization supports hyper-scale data centers by enabling Layer 2 traffic over Layer 3 overlays and simplifying virtual-machine deployment and migration.It can also provide tenants isolated views across physical and virtual infrastructure, local and remote facilities, and public and private clouds.
  • 2.2 SDN Benefits: SDN provides a programmable platform for experimenting with novel protocols and policies while separating test flows from production flows.This supports experimentation using production traffic and facilitates adoption of newer networking ideas.
  • 2.3 OpenFlow Architecture: OpenFlow is the standard southbound interface that provides software access to switch flow tables and enables administrators to change traffic handling.OpenFlow switches may be OpenFlow-only or hybrid, supporting both OpenFlow and traditional Ethernet pipelines.
  • 2.3 OpenFlow Architecture: OpenFlow architectures use switches with flow and group tables, secure controller channels, and controllers that distribute policies and instructions.Controllers may use centralized, distributed, or multi-layer structures.

3. NETWORK VIRTUALIZATION

Network virtualization decouples and isolates multiple virtual networks from shared physical hardware, enabling independent network services, policies, and operating environments. The section distinguishes NV from NFV and SDN while describing how these technologies can be combined.

  • Network virtualization decouples isolated virtual networks from underlying hardware while allowing them to share the same physical infrastructure.The physical network is used for packet forwarding, while virtual networks are overlaid with features and guarantees resembling physical networks.
  • Virtual network components include virtual servers, virtual nodes, and virtual links supporting service access, dynamic changes, and representations of physical network elements.Virtual servers implement virtual machines and can switch transparently between them as client needs change.
  • Network virtualization supports coexistence of dissimilar networks, isolated production and test domains, and deployment of distinct services on separate virtual networks.Isolation allows incompatible routing protocols and malfunctioning experiments to remain separated from production traffic.
  • NV configures packet handling, QoS, and security policies to categorize networks according to services, users, and applications.This provides independent and diverse network provisioning.
  • Dynamic mapping of virtual nodes to the physical substrate improves hardware utilization, while agile integration supports migration from legacy to advanced networks.The described approach can reduce hardware costs and support incremental network evolution.
  • NV and NFV virtualize network connectivity and functions respectively, whereas SDN modifies the physical network by separating control and data layers.SDN and NFV are complementary and independent, but can be merged to address legacy-network challenges.

4. SDN APPLICATIONS

SDN is presented as a response to data-center and Network as a Service challenges arising from scale, virtualization, heterogeneous infrastructure, and growing traffic. The surveyed applications use centralized control, OpenFlow, virtualization, and optimization to improve manageability and performance, while adoption remains constrained by reliability, scalability, latency, and security challenges.

  • 4.1. Motivation: Data-center networks face rapid growth, virtualization, rising internal traffic, and increasing scale that strain traditional networking equipment and IT operations.The paper highlights data-center consolidation, server virtualization, and widespread web applications as major change drivers.
  • 4.1. Motivation: Legacy data-center designs struggle with VM-compatible segmentation, fixed-path traffic engineering, and device-by-device provisioning across heterogeneous management systems.Hard-wired VLAN, subnet, ACL, and firewall controls are described as incompatible with VM workloads and cloud platforms.
  • 4.1.1. SDN Deployment in DCNs: OpenFlow-based SDN centrally configures virtual segments, directs flows to security services, and provisions newly added devices with network policies.The controller turns core and aggregation devices into a high-speed transport backplane.
  • 4.1.2. SDN Deployment in DCNs: 30% latency improvements were reported for the Scissor framework, while maximum evaluated power gains were 20% because no scissor operations occurred in the rack carrying 75% of DCN traffic.The power result is explicitly qualified as a best-case scenario.
  • 4.1.2. SDN Deployment in DCNs: SDN-based inter-DCN connectivity represents heterogeneous architectures through OpenFlow rules and supports VM migration between schemes without connectivity interruption.Re-routing mechanisms maintain connectivity during live migration.
  • 4.2. SDN Applications: SDN-enabled NaaS architectures provide programmable resource management, service composition, graph-based network queries, and application-specific packet processing for cloud tenants.Cloud-NaaS includes resource pooling, standardized management APIs, runtime allocation, billing, reliability assurance, and protocol services.
  • 4.2. SDN Applications: CloudNaaS optimization techniques improved traffic handling during VM placement and forwarding-entry aggregation, with good performance as provisioning requests increased.The implementation was designed to mitigate hardware limitations and use cloud resources effectively.
  • 4.2.3. NaaS and SDN Integration: Overall SDN adoption remains challenged by reliability, scalability, latency, and security obstacles despite applications in data-center networks and NaaS.The paper surveys causes and proposed solutions for these challenges.

5. SDN CHALLENGES AND EXISTING SOLUTIONS

SDN introduces challenges involving availability, scalability, latency, visibility, hardware datapaths, programmability, controller placement, and security. The paper surveys causes and existing solutions, including distributed recovery, proactive setup, architectural optimization, and placement algorithms.

  • Reliability: Controller failure can create a single point of failure, motivating distributed load-balancing multipathing that updates switches when path loads or congestion change.The described algorithm operates after controller failure and responds to traffic congestion and load imbalance.
  • Scalability: SDN controllers can become bottlenecks as switches, hosts, bandwidth, and flows increase, with current capacity far below large-data-center flow demand.NOX handles 30K requests/s, while a 2-million-VM data center may generate 20 million flows/s and current controllers support approximately 10^5 flows/s in the optimal case.
  • Performance under Latency Constraints: Reactive flow setup adds controller-processing and switch-update delays, whereas proactive setup has negligible packet-arrival delay and avoids controller flow-rate limits.Reactive setup occurs after a packet misses the switch flow table; proactive setup preinstalls possible flows.
  • Performance under Latency Constraints: 50-fold controller-latency reduction was reported for McNettle, while Maestro tunes batching and pending-packet thresholds to increase throughput and decrease delay.The relevant Maestro parameters are the input batching threshold and pending raw-packet threshold.
  • Controlling the Data Path between the ASIC and the CPU: ASIC-to-CPU bandwidth limits control-path performance, while OpenFlow counters consume 24 bytes per flow-table entry and compete with lookup-table silicon.A ProCurve 5406lz measured 300 GB/sec ASIC bandwidth but 35 MB/sec ASIC-to-CPU loopback bandwidth and 10 MB/sec flow-setup payload bandwidth.
  • Use of Low-Level Interfaces between the Controller and the Network Device: SDN policy interfaces remain low-level and event-driven, requiring programmers to coordinate device-specific rules and asynchronous switch events; high-level languages aim to reduce this burden.Flog is described as learning switch software that generates precise high-level forwarding rules from learned packet information.
  • Controller Placement Problem: Controller placement affects latency, reliability, fault tolerance, and performance, with random small-k placement producing 1.4x–1.7x the average latency of optimal placement.Wide-area deployments particularly require multiple controllers whose locations affect network metrics.
  • Security: SDN security concerns are reported by IT organizations, with 12% identifying security challenges and 31% undecided whether SDN is more or less secure.The cited studies associate security risks with insufficient integration with existing security technologies.

6. RESEARCH INITIATIVES FOR SDN

Research initiatives develop SDN architectures, cloud-network services, controllers, debugging tools, routing systems, emulators, languages, and commercial platforms. These efforts target programmable network management across data centers, wireless networks, enterprises, and campus networks.

  • Research Initiatives: SDN initiatives apply programmable networking to data-center, wireless, software-defined-radio, enterprise, and campus networks.The paper describes these initiatives as proposed SDN prototypes and related research efforts.
  • SDN Architectures: Early architectures such as 4D and Ethane established centralized policy-based approaches to network control and flow routing.Ethane uses a centralized controller holding network policies and switches that receive forwarding instructions.
  • Cloud-Network Frameworks: Meridian provides cloud users declarative and query APIs, orchestration services, topology views, routing, scheduling, and virtual-network creation over physical infrastructure.Its declarative API shapes multi-VM applications, while its query API supports topology and network-statistics requests.
  • SDN Tools and Languages: SDN tooling includes Onix for distributed controllers, Veriflow for detecting faulty application rules, Routeflow for combining commercial hardware with open-source routing, and Mininet for virtual emulation.Mininet supports SDN prototyping before deployment, but the paper notes poor high-load performance and lightweight virtualization limitations.
  • SDN Vendors: Floodlight, ProgrammableFlow, OpenFlow controllers and switches, and NICIRA’s Network Virtualization Platform represent enterprise and vendor implementations for physical and virtual networks.Floodlight supports OpenStack orchestration and both OpenFlow and non-OpenFlow networks; ProgrammableFlow uses custom switches and a controller.

7. CONCLUSIONS

SDN centralizes control-plane intelligence while retaining compatible switches, enabling programmable and agile network management. The paper connects this architecture with data-center and wide-area-network management, security efforts, and Network as a Service.

  • 7. CONCLUSIONS: SDN centralizes control-plane intelligence for Layer 2 switching and Layer 3 routing while using SDN-compatible switches.The generalized architecture consists of an SDN controller and compatible switches.
  • 7. CONCLUSIONS: SDN’s programmability and agility support strategies for simplifying data-center LAN and WAN management and making networks more secure.The passage presents these as uses being explored by academic researchers and network engineers.
  • 7. CONCLUSIONS: SDN also supports Network as a Service, which connects cloud computing with SDN while SDN manages forwarding decisions and network administration.The passage distinguishes NaaS as an Internet-based model from SDN’s forwarding and administration functions.
Loading 1406.0124v1…