Modeling and Performance Analysis of P4 Programmable Devices

Abstract

In our digitized society, emerging applications require highly performant and flexible networks that can adapt to satisfy varying connectivity needs. P4 as a domain-specific programming language for data plane pipelines introduces the required flexibility through easy-to-use programmability. However, the performance of P4-capable devices is still an open question that has not yet been completely addressed. Understanding whether a P4-enabled device can meet the performance requirements for a specific network function pipeline is key for planning as well as deployment scenarios in a communication provider network.In this paper, we propose a simple analytical model that can quickly predict the performance of network functions written in P4 for a given device. The programmable data plane of P4 devices is modeled as a forward queuing system with a variable service rate that depends on the complexity of the configured data path program. On top of the data plane model, the controller's interaction is modeled as a feedback queuing system. We evaluate the accuracy of our model through a parameter study and simulation. The evaluation reveals corner cases, which are analyzed to formalize a constraint on the service rate using model parameters to guarantee stable system performance.

Performance analysis of general P4 forwarding devices with controller feedback: Single- and multi-data plane cases

“…P4 programmable data planes extended the programmability offered by the SDN paradigm to the data plane. Similar to the approach presented in [11], the model in [9] adopts a feedback-oriented queueing system to abstract the behavior of the system. Moreover, it assumes that the processing delay, and thus the forwarding delay, at the data plane can vary based on the complexity of the loaded P4 data path.…”

“…In a first step to model the packet sojourn time for P4 devices, the authors in [5] measured the mean sojourn time of packets for the P4-enabled devices Netronome SmartNIC [6], NetF-PGA [7], and T4P4S [8], based on the complexity of the loaded P4 program. Those measurements were then further used in [9], where one P4 forwarding device with controller feedback is modeled as a Jackson network [10]. Their results allow a first understanding of the behavior and limitations of a P4 device with controller interaction.…”

“…Their results allow a first understanding of the behavior and limitations of a P4 device with controller interaction. However, as service times in [9] are assumed to be exponentially distributed, both in the data plane and the controller, it may not reflect the behavior of real devices and therefore the analytical performance results can vary from the real ones considerably.…”

The Future Roadmap of In-Vehicle Network Processing: A HW-Centric (R-)evolution

“…There is an extensive collection of works dealing with P4 in the state of the art, some of them focusing on the description of P4 packet processors [126], surveys of P4 potential applications [127], [128], [129] performance analysis [130] or extensions to support heterogeneous dataplanes [131] as well as the definition of an abstract model for programmable data planes [132]. The capabilities to exploit application layer processing are explored by authors in [133], enabling the fulfillment of FR7.…”

“…Safety and Security: Automotive Ethernet in On-Board Diagnosis & in-vehicle networking [26] Cooperation or competition? Coexistence of safety and security in next-generation Ethernet-based automotive networks [42] Secure Automotive On-Board Electronics Network Architecture [101] TSN: Assessments of Real-Time Communications over TSN Automotive Networks [58] Timely Survey of Time-Sensitive Networking: Past and Future Directions [63] DDS and DDS with TSN: DDS middleware on FlexRay network [66] Multi-Level Time-Sensitive Networking Using the Data Distribution Services for Synchronized Three-Phase Measurement Data Transfer [67] Using DDS over TSN to support NATO Generic Vehicle Architecture for Land Systems [68] SDN: OpenFlow [71] Software-Defined Networking in Automotive [72] An In-Vehicle Software Defined Network Architecture for Connected and Automated Vehicles [73] In-Vehicle Software Defined Networking: An Enabler for Data Interoperability [74] Retrofitting SDN to classical in-vehicle networks: SDN4CAN [75] An SDN Architecture for Automotive Ethernet [76] SDN with TSN: SDN-based configuration solution for IEEE 802.1 Time Sensitive Networking [62] Software-Defined Networks Supporting Time-Sensitive In-Vehicular Communication [77] Software-Defined Time Sensitive Networks Configuration and Management [78] SDN for safety applications: Fault-Tolerant Dynamic Scheduling and Routing for TSN based In-vehicle Networks [81] Dynamic Network Reconfiguration in Safety-Critical Aeronautical Systems [82] Service oriented Architecture: A distributed in-vehicle service architecture using dynamically created web Services [83] SODA: Service-Oriented Architecture for Runtime Adaptive Driver Assistance Systems [84] Programmable Dataplane:A Survey on Data Plane Programming with P4[127] An Exhaustive Survey on P4 Programmable Data Plane Switches: Taxonomy, Applications, Challenges, and Future Trend[128] The programmable Data Plane: Abstractions, architectures, algorithms, and applications[129] Modeling and Performance Analysis of P4 Programmable Devices[130] P4 to FPGA-A Fast Approach for Generating Efficient Network Processors[135] One for All, All for One: A Heterogeneous Data Plane for Flexible P4 Processing[149] …”