Ethernet Network Processor: Architectural Innovations for SDN Implementation in Huawei S12700 Agile Switches

Ethernet Network Processor: Architectural Innovations for SDN Implementation in Huawei S12700 Agile Switches

Many metrics such as bandwidth, latency, manageability, and security are used to evaluate the performance of network devices. Among other indicators, the most important is forwarding technology. ASIC (Application Specific Integrated Circuit) microcircuits allowed increasing the capacity of switches from 10/100 Mbit / s to N Gbit / s and N x 10 Gbit / s, however today new e-commerce services, online media, BYOD, continue to increase traffic volumes, requirements to functionality and quality guarantees. The SDN (Software Defined Networking) architecture, designed to control data flows and integrate with management systems for the interaction of the network, users and applications, is called upon to become an effective solution for modern IT tasks.
In this article, we’ll talk about the architectural design of Huawei Enterprise — the Ethernet ENP (Ethernet Network Processor), designed to increase performance when implementing the SDN architecture. The ENP solution became the technological basis for the functionality of the Sx700 Alige switches, the first of which was the S12700. In the future, Huawei plans to use ENP in the S9700, S7700 and lower models of switches.
Tools for developing new functionality include ASICs and NP (Network Processors). The advantage of ASIC in speed and reliability, the advantage of NP in flexibility and programmability. However, with high performance, ASIC chips only support predefined service algorithms and a fixed set of protocols. Therefore, the most significant problem in ASIC technology is the inability to flexibly expand functionality. The example below shows a fixed hardware forwarding scheme in L3 switches. Adding a new algorithm, such as MPLS or SDN, requires chain adjustment, however, changing the hardware design of the chip may take years.



Commercial NPs are designed as an alternative to ASIC for implementing complex schemes of software traffic processing, security and quality of service - for example, L4-L7 analysis, cryptography, queue management. Instead of ASIC hardware logic, NPs offer software-configurable computing resources for quickly deploying new applications.
Commercial NPs divide tasks between separate Network Processing Unit (NPUs), a collection of processors that are brought together to increase productivity. Each processor group has a dedicated memory area for instructions describing algorithms for performing tasks using program code, which makes it possible to implement new network protocols and data processing scenarios without the need to change the hardware design of microcircuits. For example, IPv6 and the corresponding route tables, new types of tunneling, or traffic encryption can be implemented in NP by changing the microcode, while using ASIC technology will require replacement of equipment. At the same time, improper resource planning can lead to downtime of some and overloading of other NPUs, leading to bottlenecks and degradation of productivity. Another disadvantage of NP is the limited instruction size and performance of the NPU group, especially when it is necessary to implement more demanding tasks than was planned during the NP development.

Combining a set of processors on a single platform makes it possible to increase NP performance to ASIC level. In addition, the increase in data processing speed is provided due to hardware increase in the clock frequency of the chip. However, the flip side of increasing configuration and clock speed is the increase in power consumption.
Based on more than 20 years of R&D experience in microelectronics, Huawei has developed the next generation NP - Ethernet NP (ENP), which take into account:

• new functional requirements, in particular, the SDN architecture, which involves switching a large number of data streams and the use of separate routing tables, which are determined by the network controller using the OpenFlow protocol;
• reduced energy savings, as opposed to a significant increase in the power of devices based on commercial NP;
• removing restrictions on the volume of instructions and improving memory performance.

Unique solutions in the design of Huawei ENP are: a single instruction area, optimized Smart Memory, a transfer table for millions of records and energy saving.

Unified instruction area
In contrast to the standard implementation on commercial NPs, the ENP instruction area is not assigned to each NPU group, but a single one for all processor groups. Each of the NPU groups can execute any set of instructions - from header analysis to data modification. Thus, functionality should not be distributed between different NPU groups, and all resources are available to perform any of the tasks. Huawei solution eliminates the problem of distribution of functions for performance planning, reduces the time to develop new functionality.
In addition, ENP has several parallel data processing chains - this increases productivity and optimizes the operation of I / O interfaces. In particular:

• ENP can perform parallel processing of protocols of different levels L2, L3, MPLS, VPN;
• Due to the fact that the size of instructions for NPU groups in commercial NPs is limited, complex services were distributed to several NPU groups that performed their tasks sequentially; a single area of ​​instructions in ENP allows you to perform service on one NPU group, since the size of instructions can be significantly increased.

Smart Memory
Both ASIC microprocessors and most commercial NPs share data processing and storage functions. Interaction with memory elements creates a delay, and execution in separate modules increases energy consumption. Therefore, in its solution, Huawei integrated Smart Memory elements into ENP and thereby optimized the speed of data exchange.
Smart Memory combines the functionality of storing routing tables and finding routes, and is also able to process traffic, in particular, fulfilling data rate limits or participating in statistics collection.
Millions of Forwarding Entries
Huawei Enterprise’s hybrid SDN solution uses the OpenFlow layer and the traditional routing layer to control data transfer in parallel. This approach allows for smooth migration to the SDN architecture, without disrupting the operation of existing services. A similar solution is used by other market participants, however, the built-in flow tables in ASIC solutions are small. One of the industrial solutions is the use of external TCAM modules (Ternary Content Addressable Memory), which combine the forwarding table, QoS and packet filters. However, such solutions are also limited in speed and table size.
The ENP solution supports up to 16 million data streams, providing flexible service management capabilities for tens of thousands of campus employees.

Energy Saving The
power consumption of ASICs is fixed and usually lower than commercial NPs. The ENP solution reduces power consumption depending on the amount of traffic and processing complexity, since ENP can block unused NPU groups. As traffic increases, ENP activates resources for several nanoseconds. The success of this mechanism has been tested by test laboratories, which recorded the ability of Huawei switches to cope with a sharp surge in traffic.
The use of energy-saving technology leads to a reduction in power consumption by 30%.
The power consumption of the chip is directly proportional to the clock frequency. Therefore, in the ENP solution, Huawei integrated speedometer, voltage and synchronization controllers. The speedometer is on the way to data NPU groups to monitor the volume of traffic. Changes in the data rate lead to changes in the clock frequency of the data channel. At the same time, changes in the number of packets lead to an adjustment in the number of active packet processors. If the NPU group is blocked, then the processors are not supplied with power and synchronization. If necessary, processors instantly turn on.


ENP applications
Huawei Enterprise uses ENP processors in the 1GE and 10GE interface modules, which are recommended for connecting access level aggregating switches and implementing advanced functionality. At the same time, the 10GE, 40GE, and 100GE trunk interfaces are formed by modules using the ASIC architecture to achieve maximum performance and MPLS support.
To preserve the health of existing applications and network solutions, Huawei Enterprise implemented a solution called “OpenFlow Hybrid”, which simultaneously provides the following:

• switching and routing functions (3 million IPv4 routing records, 1 million IPv6, 128 thousand IP multicast, 1 million MAC and 256 thousand packet filters),
• SDN for managing data streams (16 million streams).

The SDN implementation in the Huawei POF (Protocol Oblivious Forwarding) solution is based on the OpenFlow standard and is backward compatible with it for interoperability with other manufacturers. POF technology is distinguished by the ability to work with any protocols and non-standard packages, while the OpenFlow standard involves only operations with IP. ENP processors expand Huawei's ability to work with new protocols that can be implemented programmatically, while ASIC-based SDN implementations deprive manufacturers of flexibility.
A practical example of using the SDN of Huawei Enterprise is encapsulating data and adding custom information to the packet headers. In the test scheme assembled by Miercom, the formation of a non-standard package containing information about the location of the sender was verified. Further information was later used to identify the data stream and direct it over the network.
In addition to SDN, ENP-based interface modules support advanced functionality for aggregating wireless access traffic, implement quality control and localization of accidents. The subscriber aggregation solution includes a distributed access controller AC (Access Controller) and a remote access server BRAS (Broadband Remote Access Server), which turns the Huawei S12700 switch into a high-performance WiFi traffic hub that can combine up to 4 thousand access points AP (Access Point) serving up to 65 thousand simultaneous connections. The total throughput of the solution is 800 Gb / s. For the convenience of managing the AP and remote ports of aggregating switches, Huawei's Agile solution implements the technology of unifying virtual factory SVF (Super Virtual Fabric).
Integration of subscriber aggregation and SDN solutions allows implementing user service policies and, if necessary, operating vertically integrated applications with traffic, allocating it to virtual secure access environments. At the same time, the concept of data flows, implemented in the Huawei S12700 switches, allows you to centrally manage quality throughout the network. Packet Conservation Algorithm for Internet (IPCA) technology of isolated network streams allows you to identify the signs of priority traffic on the campus controller once, and then transfer the requirements to all network devices. ENP-based solutions identify priority traffic flows and control how it flows through the network at each transit node. And for quality control and localization of accidents,

The iPCA implementation is another example of Huawei’s technological leadership and practical application of SDN.