Recent work by different groups seek to address infrastructure issues of AI, analytics, and other data-intensive workloads.
The large volumes of data businesses routinely use for analytics and to train artificial intelligence (AI) models is overwhelming hyperscaler and enterprise compute and storage resources. Simply put, their infrastructures do not meet the needs of modern data-intensive applications.
Just last week, Cisco, collaborating with NVIDIA, Intel, AMD, NetApp, Nutanix, Pure Storage, and Red Hat, sought to address the issue with the announcement of Cisco Validated Designs (CVDs) for AI use cases.
In another effort, the Ultra Ethernet Consortium recently formed with open-source foundations and leading technology companies as founding members. The goal of the group is to improve one infrastructure aspect, namely ethernet delivery speeds for high-performance networking,
The group stresses how the advancement of artificial intelligence research over the past few years has pushed hardware, software, and system providers to the limit of their capabilities, with new hardware and systems built specifically to meet the demands of these research projects.
The consortium aims to improve the architecture for high-performance tasks, such as artificial intelligence and machine learning research and development. It also aims to improve the Ethernet communication stack architecture, while maintaining the ubiquity of flexibility of Ethernet to handle workloads at scale.
“This isn’t about overhauling Ethernet,” said Dr. J Metz, chair of the Ultra Ethernet Consortium. “It’s about tuning Ethernet to improve efficiency for workloads with specific performance requirements. We’re looking at every layer – from the physical all the way through the software layers – to find the best way to improve efficiency and performance at scale.”
Ethernet is one of the key building blocks of the internet and the web, and there is a keen focus by The Linux Foundation which is hosting this consortium and others to maintain the authenticity and open-source nature of Ethernet, instead of creating a new format.
Some of the key goals for the consortium are to develop specifications, APIs, and source code for each layer of the Ethernet. This includes electrical and optical signaling characteristics for Ethernet communications, extending the link-level and end-to-end network transport protocols, reducing end-to-end congestion and telemetry signaling, and storage, management, and security constructs for a wide variety of high-performance workloads.
Some of the other founding members include AMD, Broadcom, Cisco, Hewlett-Packard Enterprise, Intel, Meta, and Microsoft. Notable absentees are Google and Amazon, responsible for a large amount of internet traffic and artificial intelligence research.
A final word: Infrastructure is the key
The work of these two efforts (i.e., the consortium and Cisco and its partners) shows the need to address infrastructure issues for AI, analytics, and other data-intensive workloads to run optimally.