SPDX-License-Identifier: Apache-2.0     
Copyright (c) 2020 Intel Corporation

O-RAN Front Haul Sample Application in OpenNESS

Introduction

Recent and incoming telecommunication standards for Radio Access Network (RAN) tend to introduce open network interfaces that are expected to become adopted by broad numbers of RAN vendors and operators. Networks based on common standards are thought to be more inclined to innovations. Thus, standardization committees aim to support the global industry vision and encourage emerging multi-vendor, interoperable, and innovative virtualized RAN (vRAN) to enable vRAN shift to the Cloud and exploit the opportunities the Cloud has to offer—scalability, efficiency, cost reduction, and more. Flexible Radio Access Network (FlexRAN), which is part of proof-of-concept work at Intel, demonstrates vRAN deployment on Intel® architecture. It follows the most recent RAN standards and deploys innovative software and hardware solutions proposed by Intel to refine the baseband L1 functionality. Recently, 5GNR FlexRAN has started supporting the open Front Haul* interface standard introduced by the O-RAN Alliance* here.

The focus of this white paper is to show how OpenNESS facilitates the deployment of 5GNR FlexRAN Front Haul functional units based on O-RAN specifications at the Network Edge. It also demonstrates how OpenNESS may assist in exploiting the capabilities of the X700 family NICs to address the challenges related to 5G RAN evolution including fast-growing user traffic and the move towards the Edge Cloud. This document describes the Intel® Ethernet Controller X710 new capability known as Dynamic Device Personalization (DDP). It provides the steps for utilizing this feature on the OpenNESS platforms. DDP technology has been previously implemented and tested within LTE FlexRAN L1 and proven to reduce network latency and the number of CPU cycles used for packet processing, leading to the increase of the overall network throughput. Choosing DDP is a promising option for removing the network bottleneck related to packet filtering and realizing stringent latency and throughput requirements imposed onto 5G networks. Tests performed with FlexRAN using LTE Front Haul interface based on Ferry Bridge (FB), the codename of a technology from Intel, and incorporating the DDP capability of Intel® Ethernet Controller X710 showed up to 34% reduction in CPU cycles used for packet processing. Whereas tests performed on Multi-access Edge Computing (MEC) solution demonstrated a nearly 60% reduction in network latency. These findings are already described in an incoming white paper “Dynamic Device Personalization: Intel Ethernet Controller 700 Series - RadioFH Profile Application Note”. Shifting towards DDP for increased performance is also a promising option for the Network Edge. Such deployment has already been tested on Kubernetes* architecture and described here.

Dynamic Device Personalization

In contrast to the Intel® Ethernet Controller X550 static pipeline, the packet processing pipeline built into Intel® Ethernet Controller X710 is programmable. It can be customized to achieve different packet filtering on separate NICs to meet the requirements of a variety of customers, or it is possible to enable different protocols and classification types (new protocols/filters ) for separate network segments. As a result, 64 unique classification configurations are possible on Intel 700 Series NICs. Intel NIC XL710 supports up to 64 PCTYPEs and up to 192 PTYPEs. However, only 21 PCTYPEs and a few PTYPEs are defined in the default profile. The configuration options are limited because of the limited resources on the hardware itself. Each PCTYPE needs its own filters’ configuration space. Enabling new configurations by overwriting the existing PCTYPES can be achieved through a firmware update. However, DDP capability provides a means to replace the original device configuration profile in run time using a DDP profile. A DDP profile is a binary file that loads new PTYPEs and PCTYPEs into the hardware configuration space. Then, each new PCTYPE needs to be mapped to an unused DPDK flow type. The original profile is saved and may be restored on the device in runtime as well. Some packets may be lost during the profiles switching. During the processing of incoming traffic, the NIC’s internal parser, filters, and switch are used to perform on-NIC packet classification and steer the traffic to specific queues from where the software application can take them over for further processing defined per queue. To classify new packet types (xRAN) and protocols (eCPRI) on the Intel® Ethernet Controller X710 pipeline, they must be enabled on the NIC through device reconfiguration. Normally, this requires a firmware upgrade involving a cold restart of the server. But with DDP, new packet classification types can be added in runtime, and there is no need for a firmware upgrade requiring machine power off. Intel® Ethernet Controller X710 fully supports DDP since Firmware Version 6.0.

X710 Flexible Pipeline

The flexible pipeline added onto the XL710 NIC consists of four main functional units:

  • buffer
  • parser
  • switch
  • classifier filter: Flow Director (FD) and Hash Filter (Receive Side Scaling (RSS)

Before the buffered packet gets to the filters block, its headers are parsed, and specific header fields are extracted into a Field Vector (a metadata buffer that is attached to the packet). The parser detects the packet type (PTYPE), which is saved to the RX descriptor and defines the packet classification type (PCTYPE). Some fields from the Field Vector are used in the next functional block for switching the packet to the proper PF or VF. If FD and hash filters such as RSS are enabled for the pipeline, they are used to select the packet’s destination queue within the PF/VF it had been switched to. The information stored in the Field Vector fields is used by the filters to steer packet destination queues to where the host expects them. The Intel® Ethernet Controller X710 provides programs with advanced classification and traffic steering functionality due to the following features of its pipeline:

  • Multiple queues are available per virtual function (VF) / physical function (PF).
    • 1536 queue pairs in the device, 16 queues per VF and 128 VFs – 384 Virtual Station Interfaces (VSIs) in total
  • Use of receive-side parser for packet classification.
  • Switch and filter blocks use RSS and/or FD.
  • The received packets are buffered in a packet buffer.
  • The parser inspects different header fields such as Ethernet, IP, and UDP.
  • The switch module switches the VSI/PCI function using parsed header fields.
  • The flow rules of the switch are programmable by the upper layer software.
  • The packet filter/classifier module implements the queue selection logic.
  • FD supports exact 8K match rules and wild card rules.
  • An 8K exact match rule can use up to 16 flexible bytes, along with L2, L3, and L4 header fields.

The diagram below depicts the functional units of the Intel® Ethernet Controller X710 pipeline: Intel X710 NIC Pipeline

Flow Director

Flow Director (FD) is a packet filtering functionality provided by DPDK for routing packets to specific queues on NIC ports. DPDK Flow Types are equivalents of PCTYPEs used by the hardware. FD uses Flow Types, specific input set per Flow Type (exact match of the input set’s selected fields -tuples), or flexible payload fields to match packets and direct them accordingly to existing FD rules. A flow rule consists of a set of attributes with a matching pattern and a list of actions. Its primary use is to match actions with Flow Types. Actions that can be performed on the matched packets include dropping, diverting to a specific queue, port, or a virtual/physical device function, performing tunnel offloads, and marking packets. FD uses the Toeplitz hash function to calculate the hash signature of the input set generated for each filtered packet.

Received Side Scaling

Received Side Scaling (RSS) is a hash filter used to statistically distribute packets into queues belonging to a queue region by maintaining load balance. The queues are assigned to different processors. Thus, the packet processing work is shared among several processors. RSS calculates the 32-bit hash signature out of the input set selected from the Field Vector. The 32-bit hash function based on the Toeplitz algorithm or an XOR scheme is used for generating that hash. Based on its hash, value flows are assigned to a queue region, including one queue at least. Packets of the same flow will be distributed evenly to all queues within the queue region. The flow chart below shows an example of using FD and RSS in the pipeline to achieve advanced classification and lookup.

Advanced classification and lookup on XL710

DPDK support

Special Admin Queue (AQ) commands have been added to the Intel® Ethernet Controller X710 firmware version 6 to assist new profile addition/removal to/from the hardware, and to retrieve information about the profiles currently registered on the NIC. The new AQ commands can update NIC configuration tables or read/retrieve their content. DPDK started supporting the DDP feature by providing software calls to the AQ commands gradually through versions 17.05, 17.08, and 17.11. Since then, DPDK provides full support for working with DDP profiles and enables the recognition of protocols not known to DPDK flows (rte_flows). Currently, DPDK APIs can be used to load/remove profiles and retrieve a list of registered profiles and their descriptions. DPDK “testpmd” application deploys RTE APIs to manipulate the DDP profiles and uses them to configure/enhance packet filtering on the NIC.

More information on Intel® Ethernet Controller X710 flexible pipeline can be found in the [here] (https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xl710-10-40-controller-datasheet.pdf).

O-RAN Specification for vRAN Front Haul

According to the O-RAN Alliance Fronthaul specification, the physical layer of the 5GNR vRAN network is divided into two functional units: O-RU and O-DU. The functional split of the eNG/gNB functions aims to keep the O-RU simple. O-RAN selected two possible options for the split called 7-2x: Category A (lower split) and Category B (upper split). The difference between the two split categories refers to precoding placement. In category A, precoding is done on the O-DU side. In Category B, it is done on O-RU. The O-RAN Alliance specifies the packet formats for the data flows over the 5GNR O-RAN Front Haul.

The image below presents the 7-2x functional split and places Precoding on O-DU for Cat A split and on O-RU for Cat B split: O-RAN Font Haul Functional Split 7.2x

Front Haul Data Flows

The following lists supported data flows in the 5GNR Front Haul defined by O-RAN.

  1. User Plane
    • Data Flow 1a: Flows of IQ Data in FFT frequency domain on DL
    • Data Flow 1b: Flows of IQ Data in FFT frequency domain on UL
    • Data Flow 1c: Flow of PRACH IQ data in FFT frequency domain
  2. Control Plane
    • Data Flow 2a: Scheduling commands (DL and UL) & Beamforming commands
    • Data Flow 2b: LAA Listen-Before-Talk (LBT) configuration commands and requests
    • Data Flow 2c: LAA LBT status and response messages
  3. Synchronization-Plane

O-RAN specification includes Management Plane. Examples of the responsibilities of M-plane are:

  • Generic static O-RU configuration
  • O-RU management
  • Compression configuration
  • Choosing a power-saving setting

The following diagram presents the data flows between the proposed O-DU and O-RU:

O-DU - O-RU Data Flows

U-Plane packets exchange IQ data between O-RU and O-DU. The data is transmitted symbol by symbol within U-Plane messages. The main purpose of the C-Plane messages is to transmit data control information required for the processing of user data. The data-associated control information is typically sent to every slot (or every TTI in LTE) in a different data flow with its header and encapsulated payload (C-Plane messages). C-Plane messages send information regarding the next expected U-Plane data packets.

An example of DL C-Plane and DL U-Plane Package Exchange is shown in the following diagram:

DL C-Plane and DL U-Plane Package Exchange

Information exchanged between O-DU and O-RU is encapsulated into standard Ethernet frames. Ethernet is used as a transport mechanism for both U-plane and C-plane. M-Plane should be responsible for MAC address resolution and VLAN tagging of the frames. According to the O-RAN standard, the Ethernet header is followed by an eCPRI transport header. Additionally, the eCPRI header can be encapsulated into IP/UDP datagrams but this solution is not supported by 5GNR FlexRAN.

Native Ethernet Frame with VLAN:

Native Ethernet Frame with VLAN

Native IPv4 Packet with VLAN:

Native IPv4 Packet with VLAN

eCPRI transport header:

eCPRI Transport Header

The “ecpriMessage” field uses different values for U-Plane (0), C-Plane (2), and eCPRI One-Way Delay Measurement message (5), which is not supported in 5GNR FlexRAN. The standard proposes that U-Plane and C-Plane transmit data over separate channels.

U-Plane Packet Structure

Following the transport layer in the U-Plane packets, there is the Radio Application Header used for time reference:

Radio Application Header

The Radio Application Header is followed by repeated Data Sections that start with the Application Header that is followed by the data in the form of multiple PRBs. Each PRB is built of 12 IQ samples.

The header of the Repeated Data Section:

Section Application Header

Data Section Payload Format:

Section Payload

Currently, Section Type 1, 3, 5, and 6 are defined for U-Plane packets. All of them have the same field structure.

C-Plane Packet Structure

eCPRI packets with message type 2 (C-Plane) can contain different section types that apply to the specific purpose of the control plane packet:

  • Section Type 0 Unused Resource Blocks or symbols in Downlink or Uplink
  • Section Type 1 Most DL/UL radio channels
  • Section Type 3 PRACH and mixed-numerology channels
  • Section Type 5 UE scheduling information (UE-ID assignment to section)
  • Section Type 6 Channel information
  • Section Type 7 LAA

Following the eCPRI transport header, C-Plane messages have a Radio Application Common Header with the number and type of the Radio Application Sections included in the packet, followed by the actual sections. The sections send information applicable to specific PRBs sent over the U-Plane. Sections of different Section Type have different fields structure.

Radio Application Common Header of a C-Plane Packet Section Type 1: