OCP China Day: accelerating community-driven industrial open-source innovation
Inspur hosted OCP China Day in Beijing on November 13 as part of the larger overarching annual OCP Tech Week, attracting more than 600 IT engineers and data center operators to the conference. Ecosystem leaders and OCP members, including Facebook, Intel, Microsoft, Baidu, Tencent, Alibaba, NVIDIA, Nokia, and Seagate, shared latest developments and innovative practices in open computing technology.
The widespread application and rapid development of 5G and AI are spearheading new initiatives for the open computing community: OAI for AI acceleration, OpenEdge for 5G edge computing, and SONiC for open network standards are becoming the key directions for new industrial applications of open computing. This marks the expansion of OCP into emerging technologies like 5G, edge computing, and AI, beyond the initial niche focus on data center infrastructure. Today, the OCP community has more than 200 members, with 8 technical project groups including data center infrastructure, servers, storage, hardware management, and open acceleration hardware. Following OCP’s growth, OCP China Day has become one of the world’s largest open computing forums.
All-in-one: rack-scale deployment as the new normal
Currently, a dominant online economy is driving the massive growth of hyperscale data centers, and the rapid deployment of whole racks, intelligent operation and maintenance, and flexible expansion has become the new normal. Rack-scale open technology standards like Open Rack, Scorpio, and Open19 have been quickly adopted by large data centers, for the benefits they bring of high density, easy maintenance, cost efficiency and flexibility. Such benefits were made evident when Inspur deployed over 10,000 nodes of L11 Scorpio racks in 8 hours to Baidu’s data center in 2019.
The OCP community also began OpenRMC, a data center automation and maintenance technology suite, to meet increasing requirements by large centralized data centers for IT facility management, rapid equipment configuration, remote fault location and automation, and intelligent operation and maintenance. So far, the project has launched the I9000 hardware verification of the OpenRMC code framework, and plans to release version 3.0 in Q4 of 2021.
OAI: accelerating open innovation in AI
OCP’s development of OAI open standards has greatly catalyzed the integration and innovation of AI computing infrastructure, an area characterized by hardware complexity and ecosystem fragmentation. The project’s technical focuses include OAI, OAM, UBB, and HIB, addressing structural design, temperature, management, power supply, hardware security, and availability. The aim is to establish a set of technical standards compatible with various AI accelerators to simplify the development of AI hardware infrastructure.
This year, Inspur collaborated with Intel Habana and Cambrian to complete the adaptation of Inspur’s OAI system, which includes a UBB SPEC design specification based on the OAM standard to enhance the compatibility of various OAM-compliant AI accelerator chips.
Open edge computing: enabling large-scale application
Recognizing the growing demands for edge computing brought on by 5G and AI, OCP and ODCC set up OpenEdge and OTII project teams respectively. OpenEdge went on to develop the world’s first universal edge server standard, solving the fundamental problem of integrating edge server and telecommunications specifications. Meanwhile, the OTII project team, consisting of China Telecom and Intel, among others, launched its first product in 2019 which saw large-scale applications.
More open edge projects followed: China Mobile deployed 105 edge pilot projects and more than 200 industry applications; China Mobile and China Unicom conducted UPF tests and edge pilots based on OTII servers, and continued to improve OpenUPF for 5G and edge computing.
As a member of both projects, Inspur has developed various edge computing solutions such as edge micro data centers, edge cloud servers, and edge AI servers in industries like mining, oil and gas, IoT, and intelligent cities.
SONiC: open sourcing the data center network
In order to meet the rapid growth of AI, data analysis and high-performance computing, data centers are transitioning to software-defined infrastructure. SONiC (Software for Open Networking in the Cloud) was formed to answer to the increasing need for intelligence, automation, high-performance, and flexibility in the data center network transmission layer. As a sub-project of the OCP Open Network project team, SONiC has delivered about 15 new feature sets in the past six months –including functions like D-BUS for container and host communication, dynamic port splitting, and port mirroring– with plans to deliver 35 more features and enhancements in the next half year.
SONiC members have so far contributed a great deal to the community. NVIDIA improved the accuracy and efficiency of the Spectrum switch network fault diagnosis and troubleshooting through SONiC. Inspur has conducted testing, security reinforcement and application scenarios to support the flexible SPINE-LEAF network architecture, which is more reliable than the traditional 3-layer network architecture. NOS, developed by Tencent, exhibits the architectural advantages of SONiC and provides ease of development and maintenance of switches through its mature tools and ecosystem.
SONiC’s future ambitions include features like support for 400G networks, Chassis scenarios, Kubernetes management, high-speed RDMA, and security modules, to empower the industry.
Open source technology and the continuous integration of open source software and hardware is profoundly changing the global IT industry’s model for collaboration. As the world’s largest open computing community, OCP brings together the world’s leading companies to enhance dissemination of leading technologies and accelerate the evolution of data centers, expanding cutting-edge technologies beyond the hyperscale data center to drive industry development.