In mid-2023, major networking companies including Cisco, Arista, HPE, and Intel established the Ultra Ethernet Consortium (UEC) with the aim of enhancing Ethernet technology to better cater to the demands of artificial intelligence (AI) and high-performance computing (HPC). Nearly two years later, they unveiled the inaugural UEC 1.0 specification, which incorporates various improvements aimed at achieving low-latency transport for high-throughput networking, along with a modern Remote Direct Memory Access (RDMA) strategy, transport protocols, and congestion control mechanisms.
J Metz, the chair of the UEC steering committee, noted that the original focus on HPC shifted due to the rapid rise of AI, following the advent of ChatGPT. This transformation reinforced the consortium’s mission to address a broader market need as AI gained popularity.
Metz considers the 1.0 release a pivotal milestone in the networking landscape, signifying that a collaborative effort among several companies had matured into a stable and reliable standard. He emphasized that while Ethernet has been predominantly versatile, it was often perceived as insufficient for more demanding applications. However, UEC’s approach demonstrated that tuning Ethernet to meet specific workload needs, though complex, is achievable.
A core advancement in the UEC 1.0 specification is the revised congestion control mechanism, tailored specifically for AI workloads. This new method moves away from the traditional reliance on lossless networks, allowing receivers to actively manage sender transmissions. Such capabilities are essential in constructing larger, more efficient networks, ultimately reducing training and inference times for AI operations.
Historically, Infiniband has been viewed as superior to Ethernet for HPC and AI applications due to its performance advantages. However, Metz asserted that UEC’s workload-centric approach narrows this gap, adjusting packet delivery to meet specific networking requirements without necessitating changes to the applications themselves.
The support for UEC has broadened over recent years, with various networking vendors backing the initiative. Arista Networks highlighted the significance of the 1.0 release as it outlines a new transport protocol designed for future AI and HPC needs, focusing on low latency and modern congestion control. Additionally, Arista plans to integrate UEC 1.0 enhancements across its Etherlink products.
Similarly, Juniper Networks has expressed enthusiasm for UEC’s capability to facilitate packet spraying and reorder functionality at the switch level, significantly boosting network utilization through open standards.
HPE has made substantial contributions to the UEC 1.0 specification, with roughly 75% of the Ultra Ethernet Transport Protocol stemming from its existing Slingshot technology, already in use in top-tier supercomputers. HPE is committed to incorporating the UEC 1.0 spec within its own networking products.
Looking ahead, Metz confirmed that UEC is just beginning its journey. Four specialized workgroups are actively developing projects to streamline UEC deployment to enhance usability and performance. They are also creating educational resources to assist networking professionals in integrating UEC technologies while forging partnerships with industry entities to address various networking layers.
Welcome to DediRock, your trusted partner in high-performance hosting solutions. At DediRock, we specialize in providing dedicated servers, VPS hosting, and cloud services tailored to meet the unique needs of businesses and individuals alike. Our mission is to deliver reliable, scalable, and secure hosting solutions that empower our clients to achieve their digital goals. With a commitment to exceptional customer support, cutting-edge technology, and robust infrastructure, DediRock stands out as a leader in the hosting industry. Join us and experience the difference that dedicated service and unwavering reliability can make for your online presence. Launch our website.