ProgLab: Programmable labels for QoS provisioning on software defined networks

Abstract

To ensure the quality of service of an end-to-end connection, current network solutions are mostly dependable on the differentiation between different classes of traffic. The Software-defined networking (SDN) architecture has emerged to offer network programmability, giving to network operators a programmatic control over their network. In SDN, network devices are programmed in many ways, having a standard, open, and vendor-agnostic interface, e.g., OpenFlow, enabling the control plane to instruct the forwarding behavior of network devices from different vendors. In this paper, we introduce the Programmable Labels (ProgLab) approach to support traffic differentiation with QoS guarantees as a low-cost alternative built over an SDN architecture. The idea relies on the simplification of a packet-forwarding operation which relies on the remainder of a division, instead of classical table lookup method. ProgLab computes programmable label at the control plane by solving a congruence system from Residue Number System and the co-prime numbers assigned to the switches in the path of an end-to-end connection. Such label has a meaning within this network that expresses the entire route, addressing the respective traffic class at each switch’s logical queue along the path. ProgLab approach has been implemented through the P4 language and evaluated through an emulation-based evaluation. The experiments demonstrated the feasibility of ProgLab and showed its ability in providing QoS differentiation on demand.

Introduction

Recently, a research challenge has risen by the need of dealing with flows that are emerging from next generation mobile 5G networks and Industry 4.0 applications. New services will redefine the networking ecosystem that is expected to support Ultra High Definition (UHD) 4K, Augmented Reality (AR), and the Internet of Things (IoT). In particular, latency should be reduced 30–50 times, for meeting the Ultra-Reliable and Low-Latency Communication (URLLC) requirements [1].

Meeting URLLC requirements implies that end-to-end delay needs a near-deterministic guarantee and it will indeed require us to rethink the current network infrastructure. Traditionally hardware-based networking solutions such as Multi-Protocol Label Switching (MPLS) have been proposed to deliver a cost-efficient way of routing traffic in high-performance connectionless IP networks. In MPLS, Label Switch Routers (LSRs) make forwarding decisions at each hop based on label. These decisions rely on pre-established connection-oriented paths or tunnels, called Label Switch Paths (LSPs). MPLS has offered virtual circuits which enable network operators to assign a higher priority to network traffic classes. Such benefits bring a sense of traffic predictability within the network since network paths are predetermined. However, the core of its operation has remained static at the heart of service providers and data center networks, operating with opaque label-based forwarding tables.

The rise of Software-Defined Networking (SDN) [2] opens the door to the design of new network architectures and protocols. This network paradigm enables different mechanisms for routing and forwarding [3], [4], [5], both simplifying manageability and improving delay guarantees [6], [7]. However, they usually require to rewrite the packet header in every hop, (i.e., swap and pop operations) which may lead to high variability on packet forwarding latency. This growth on packet forwarding latency occurs quite often. In particular, when the number of entries in the flow tables, i.e., network states, is higher than TCAM switch storage capacity [8], [9].

Our recent proposal for SDN architecture, named RDNA [10], has advanced on the canonical SDN approach [11]. We have argued that the decoupling between edge and core tableless switches are two building blocks towards a more programmatic QoS within SDN design. OpenFlow can be seen only as an intermediate solution between these two design extremes, neither general enough for the edge and nor simple enough for the core.

In this paper, we explore the minimalist forwarding model introduced by RDNA as a fundamental concept to enhance network programmability. Instead of forwarding packets based on traditional table lookup operations, the core nodes are tableless switches that forward packets using merely remainder of the division (modulo) operations. ProgLab approach starts by solving a congruence system from Residue Number System [10] to get a label embedded into the packet header by the edge switch. The modulo of the label with a co-prime number (assigned to the switch) gives the respective traffic class at each switch’s logical queue. The ProgLab’s merit is the ability to define the switches logical queues explicitly along the path. They express the flow class at each switch and offer a new degree of control for QoS fine-grained traffic engineering. We designed ProgLab to support virtual circuits with traffic differentiation, not supported in the original RDNA architecture.

As proof-of-the-concept, we implemented and validated a ProgLab prototype over the P4 language [12]. Besides that, we conducted the emulation tests by evaluating some QoS metrics, showing the agility of programmable labels to offer QoS differentiation on-demand. Consequently, we demonstrate the feasibility of our approach. To summarize, this work has three folds bringing the following contributions:

• The proposal of a new L2 forwarding protocol that supports the processing of ProgLab packets and legacy protocols in both core and edge switches;

• The introduction of a simple and expressive routing approach which takes advantage of SDN architecture to provide QoS differentiation for networking applications;

• The implementation of ProgLab core and edge switches enabled by P4 data plane programming language. We evaluated them in a mininet-based emulation environment.

The rest of the paper is organized as follows. In Section 2, we present the ProgLab concept discussing its design, a protocol implemented in P4 with its scalability analysis. Section 3 presents ProgLab experimental implementation in emulated environment with QoS results. In Section 5 we discuss the main contributions of this work and provides some final remarks about ProgLab. Section 6 summarizes the paper and point future research directions.