As 5G draws near, many communication service providers (CoSPs) are accelerating network transformation. China Telecom, one of China’s three major telecommunication operators, is combining the needs of applications, cloud computing, pipelines, and users by driving deep integration of cloud and network to provide a complete, flexible, and scalable “integrated cloud and network” solution. However, when driving transformation and deployment of the new cloud-native service-based architecture that controls, accesses and forwards cloud, China Telecom also needs a more flexible infrastructure with higher performance to provide core support capabilities.
As a long-term technology partner of China Telecom, Intel is committed to driving the development of next-generation network technologies and helping CoSPs use x86 general-purpose servers to build SDN/NFV-based next-generation network infrastructure through its own software, hardware and technology, such as Intel® Xeon® Scalable processors, Intel® Optane™ DC SSDs, and the Data Plane Development Kit (DPDK). The new 2nd Generation Intel Xeon Scalable processor has attracted China Telecom’s attention with more cores, higher clock speed, and the new Intel® Speed Select Technology - Base Frequency (Intel® SST-BF). And its advantages in performance and functions are indispensable for higher processing and forwarding performance of each network element in the cloud-based telecommunications network architecture.
To validate the performance of the 2nd Generation Intel Xeon Scalable processors in an NFV-based network architecture and provide more valuable references for deployment at later stages, China Telecom and Intel chose key functions such as Open vSwitch* with DPDK (OVS-DPDK) and single root I/O virtualization (SR-IOV) for a series of laboratory evaluation and verification. The testing results show that the new generation of processors have effectively improved performance gains of controlling and forwarding cloud in data forwarding and brought more flexible and resilient business support capabilities to the upper-layer applications.
Implementation of cloud-network integration and establishment of a flexible and open next-generation NFV network cannot be realized without development of general-purpose high-performance data plane capabilities. The introduction of the second-generation Intel® Xeon® Scalable processor effectively helps enhance NFVi capabilities, enabling better performance of key functions such as data forwarding, thus accelerating China Telecom's progress in building a new generation of high-performance cloud-based infrastructure based on general-purpose hardware.” —Liang Ou, architect, Intelligent Network and Terminal Research Institute, China Telecom Co., Ltd
Cloud-Network Integration Brings Challenges to Infrastructure Capabilities
Implementing the “cloud transformation” of network infrastructure with the help of cloud-network integration is an effective means for China Telecom to promote digital transformation for 5G. In its view, the significance of cloud-network integration is to provide more flexible and scalable services to achieve quality improvement, cost reduction, and better efficiency. To this end, Telecom is working with partners, including Intel, to implement transformation and deployment of a new cloud-native service-based architecture for cloud controlling, accessing, and forwarding based on the NFV architecture. These cannot be done without the support from strong and reliable infrastructure capabilities.
In the NFV infrastructure, the NFVi layer, which consists of computation, storage and network resources as well as virtualization capabilities on it, is the key to high performance of upper-layer virtualized network elements such as Virtual Broadband Network Gateway (vBNG) and Virtual Customer Premises Equipment (vCPE), as shown in Figure 1.
Figure 1. NFV infrastructure
Evidence shows that China Telecom’s introduction of many advanced hardware and software products and technologies, such as Intel Xeon Scalable processors, Intel Optane DC SSDs, and DPDK, to the NFVi layer has brought a reliable boost to its construction of more efficient next-generation NFV network architecture. In particular, the introduction of 2nd Generation Intel Xeon Scalable processors will more actively promote China Telecom's process of building a new generation of cloud network infrastructure based on x86 general hardware.
Key Technology for Data Forwarding
Data services will undoubtedly dominate the future telecom business landscape. In China Telecom’s cloud-network integration process, achieving efficient data processing and forwarding performance in cloud control and forwarding has become a key task in building network infrastructure capabilities, among which OVS-DPDK and SR-IOV are attracting more and more attention as network virtualization technologies commonly used in cloud control and forwarding.
As a high-quality open source multi-layer virtual switch, OVS delivers large-scale network automation and supports a series of standard management interfaces/protocols, so it can be widely used in cloud service environments. To avoid huge resource consumption caused by kernel-based OVS in its context switching between user space and kernel space, OVS-DPDK, which combines OVS and DPDK, gets main data processing done in user space and effectively improves data forwarding performance. Currently, OVS-DPDK is playing an important role in controlling cloud and network elements such as IP Multimedia Subsystem (IMS) and other data processing and forwarding functions.
Another highly efficient virtualization solution, SR-IOV, allows virtual machines to improve performance by efficiently sharing physical PCIe* devices after they are allocated to appropriate virtual functions (VFs). As shown in Figure 2, when the data on physical ports need to be forwarded, they will be lined up in the VFs and copied into the virtual machine user space. Since these operations do not require involvement of processor cores, more processor resources can be released to participate in resource allocation. And virtual machines can also access hardware directly through the virtualized PCIe channel in the VF. This means that SR-IOV has excellent network performance, and its deployment in network elements (such as vBNG) with large data throughput requirements can greatly improve forwarding performance of the entire NFV network.
Figure 2. OVS-DPDK and SR-IOV functional structure for NFV architecture
New Members of the Intel® Xeon® Scalable Processor Family
Compared to previous generation products, the 2nd Generation Intel Xeon Scalable processor provides more computing power for the NFVi layer, and with more cores and higher clock speed, it can address more effectively the increasing and complex requirements for data processing and forwarding. On the one hand, the number of cores in the processor has increased significantly compared to previous generation processors, for example, the Intel® Xeon® Gold 6230 processor has 20 cores, while the Intel® Xeon® Gold 6252 processor has up to 24 cores. On the other hand, whether it is the OVS-DPDK technology used in controlling cloud, or the SR-IOV technology used in forwarding cloud, they both deliver higher processing performance with higher clock speed. Therefore, a variety of new technologies are employed to increase clock speed in the 2nd Generation Intel Xeon Scalable processors.
A critical contributor to this is Intel® SST-BF. First, it boosts the processor's clock speed without increasing overall power consumption. Take the Intel® Xeon® Gold 6230 processor as an example, its clock speed is 2.1 GHz, and the Intel Xeon Gold 6230N processor with the addition of Intel SST-BF has its clock speed boosted up to 2.3 GHz, while TDP remains 125 watts.
It is well known that different cores are allocated to different workloads when a multi-core processor implements tasks. If the operating frequency for key workloads can be improved, it will effectively boost the overall performance of the system. Now, select 2nd Generation Intel Xeon Scalable processor uses Intel SST-BF to adjust the operating frequency of different cores.
Figure 3. Comparison of workloads before and after using Intel® SST-BF
Take the Intel Xeon Gold 6230N processor as an example, all user workloads are allocated to the core running at 2.3 GHz before Intel SST-BF is used for clock speed adjustment, as shown in Figure 3. After Intel SST-BF is used, users can forward some of the more critical workloads, such as high-speed user plane data, to the core running at 2.7 GHz by using a processor with asymmetric frequency configuration, while other business processes are placed on the cores with clock speed set back to 2.1 GHz. This adjustment at the core frequency level not only improves processing performance of critical workloads, but also reduces overall energy consumption of the processor.
Empirical Tests Verified Excellent Application Results
To further test and validate the performance of the 2nd Generation Intel Xeon Scalable processor in the NFV network architecture, China Telecom tested features such as OVS-DPDK and SR-IOV deployed on network elements for controlling and forwarding cloud with the support of Intel. The tests selected the previous generation Intel Xeon Gold 5118 processor and Intel Xeon Gold 6130 processor, as well as Intel Xeon Gold 6230 processor and Intel Xeon Gold 6230N processor with Intel SST-BF. The configuration of each processor platform is shown in Table 1:
Table 1. Processor platform configuration for OVS-DPDK/SR-IOV tests
In the OVS-DPDK functional test, the OVS-DPDK throughput on different processor platforms was tested with 6 different frame lengths ranging from 64 bytes, 128 bytes to 1,518 bytes. The test results are shown in Figure 4. Take the 64-byte frame length test as an example, the Intel Xeon Gold 6230 processor improved OVS-DPDK throughput by about 43% compared to the previous generation Intel Xeon Gold 5118 processor. After clock speed was adjusted with Intel SST-BF (core workload speed was set at 2.7 GHz and that of other workloads was set back to 2.1 GHz), performance was improved by around 61%.1
Figure 4. OVS-DPDK performance comparison on different processor platforms
At the same time, SR-IOV throughput performance was also tested with the TestPMD* tool. TestPMD is one of the reference applications distributed with the DPDK software package. Its main purpose is to forward packets between network ports to detect the functional characteristics and performance of different network devices. In the test, six different frame lengths ranging from 64 bytes, 128 bytes to 1,518 bytes were also configured, and the SR-IOV throughput was tested on different processor platforms.
The test data shows that the addition of Intel SST-BF effectively improves the performance of SR-IOV deployed on network elements for cloud forwarding. Take the 64-byte frame length test as an example, the SR-IOV throughput of the Intel Xeon Gold 6230N processor was increased by 7.82% compared to the Intel Xeon Gold 6230 processor after clock speed was boosted to 2.3 GHz by using Intel SST-BF, as shown in Figure 5. After clock speed was adjusted (core workload speed was set at 2.7 GHz and that of other workloads was set back to 2.1 GHz), the performance was further improved by 11.09%.2
Figure 5. Comparison of SR-IOV performance (TestPMD) before and after using Intel® SST-BF
The results from another test for L3fwd (Layer 3 Packet Forwarding)/SR-IOV further prove that the addition of Intel SST-BF significantly improves forwarding performance of network elements. In the test, the two participating processors were configured to use a single-core dual-thread operating mode. The Intel Xeon Gold 6230N processor with Intel SST-BF put the core work of key businesses on the core running at 2.7 GHz, while the Intel Xeon Gold 6230 processor without Intel SST-BF works at 2.1 GHz. The test results are shown in Figure 6: with Intel SST-BF, the processor's throughput per core was increased by 19%.3
Figure 6. Comparison of SR-IOV performance (L3fwd) before and after using Intel® SST-BF
At the same time, the usage of cores on different processor platforms was compared in the test. With the same deployment of three virtual machine workloads, the Intel Xeon Gold 6230 and 6230N processors each have five remaining cores for other tasks, which plays a positive role in improving overall system performance and reducing overall power consumption, as shown in Figure 6.
Figure 7. Comparison of usage of cores on different processor platforms
With the goal of building more flexible, efficient, and resilient infrastructure capabilities for the next generation of NFV network, China Telecom, and Intel have conducted research and a number of fruitful tests, which bring China Telecom a lot of useful reference on which to build high-performance generic data planes for controlling and forwarding cloud in future NFV architectures whilst combining with enhanced capabilities of 2nd Generation Intel Xeon Scalable processors and various features of Intel SST-BF.
Next, China Telecom and Intel also plan to deepen their collaboration in more Intel® hardware and software products and technologies. For example, they have the plan to explore and test the application of Intel® Resource Director Technology (Intel® RDT) on the topic of how to effectively allocate and manage processor shared resources, enabling processors’ core resource capabilities, such as Level 3 cache and memory bandwidth, to serve critical business processes more efficiently.
With further combination of China Telecom's extensive experience in network construction and Intel's cutting-edge hardware and software products and technologies, the two companies will work together to develop stronger infrastructure capabilities for the next generation of NFV network based on advanced x86 general-purpose hardware devices and build on it the future network architecture with smarter networks, more integrated management, more agile operations and more open systems.
Advantages of China Telecom's Solution:
- The 2nd Generation Intel Xeon Scalable processor offers more cores, higher clock speed, and greater processing performance, enabling more powerful infrastructure support to China Telecom's NFV network.
- Intel SST-BF boosts the overall operating frequency of the 2nd Generation Intel® Xeon® Scalable processor without increasing power consumption, resulting in performance gains.
- Intel SST-BF adjusts the operating frequency of different cores of the 2nd Generation Intel Xeon Scalable processor, enabling key workloads to be supported by more powerful processor resources.