Qumulo has joined the Ultra Ethernet Consortium, the organisation which aims to deliver an Ethernet-based open, interoperable, high-performance, full-communications stack architecture to meet the growing network demands of AI & HPC at scale. It is the first Networked Storage supplier to join the organisation.
It has also announced a collaboration with Intel Corporation and Arista Networks to advance the state-of-the-art IT infrastructure at the intersection of networking, storage, and data management. The combination of technologies from Intel and Arista will strengthen the Qumulo Scale Anywhere Data Management platform.
Kiran Bhageshpur, Chief Technology Officer at Qumulo, Commented, “The Ultra Ethernet Consortium’s work will shape the way data flows across the network, bringing systems, storage, and networks closer together while simplifying architectures with massive performance and reliability improvements. Coupling UEC’s forthcoming enhancements with today’s development and execution with data center switching leader Arista Networks in hundreds of production real-world scale anywhere primary storage systems reshapes what is possible in data management for the enterprise.”
Working with Arista Networks, Qumulo has deployed more than an exabyte of storage across hundreds of customers connected using the Arista Networks EOS-based switching and routing systems. The 7280 and 7800 modular data centre switches have delivered the proven leaf/spine architecture that Qumulo can take advantage of. With a converged network that reduces or eliminates costly legacy storage networks.
San Diego Supercomputer Center see benefits
One customer is the San Diego Supercomputer Center. Located at the University of California, San Diego, SDSC specializes in data-intensive computing and cyberinfrastructure. It is part of the XSEDE network, facilitating interactive sharing of computing resources, data collections, and advanced research tools. SDSC utilizes on-prem supercomputers for advanced computation. It manages all aspects of big data storage and analysis, including data integration, performance modelling, data mining, and predictive analytics.
Arista provides a high bandwidth network architecture with its switches. Brian Balderston, Director of Infrastructure at the San Diego Supercomputer Center, explains, “Our UC San Diego customers require dedicated networks for storage interconnect, based on a standard ethernet infrastructure at up to 200 Gbps of provisioned performance. The Data Science Machine Learning Platform (DSMLP) runs on Qumulo, where thousands of students execute performance-sensitive AI workloads concurrently. These students require optimal configurations to ensure efficiency when running GPUs over thousands of NFS connections.
“With an Arista deep buffer leaf-and-spine and Qumulo Scale Anywhere data management, we are able to segment performance required for the DSMLP, lower costs, and deliver a common filesystem for other student organizations as well.”
Stronger together
As organisations look to deploy supercomputing requirements. It is wise to remember that the compute power needs data. That data needs storage that is high-speed and able to meet the demands of powerful computing power today. For better connectivity, those computers needs faster networking. Which is why companies such as Arista and Intel became steering members of the UEC. And why Qumulo has also joined the organisation.
Technology vendors must work together to connect their systems and create a modern architecture that is not only seamlessly integrated but also flexible and adaptable to meet the needs of today and tomorrow. Qumulo aims to provide a primary enterprise file storage solution that will enable critical business systems, SIEM systems, backups, training and inference in AI workloads, and large-scale distributed systems and applications in the data centre and cloud.
Ed Chapman, Vice President of business development and strategic alliances at Arista Networks, commented, “As we develop high-performance and scale networks for the largest AI pods on open and interoperable IP and Ethernet protocols, the evolution to Ultra Ethernet is to simplify the network and bring compute, AI processing, and storage together. Qumulo joining the UEC is further validation that Ethernet and IP are the right foundation for the next generation of general purpose, cloud, and AI computing and storage.”
Enterprise Times: What does this mean
There are other storage providers within the UEC. But Qumulo’s claim that it is the first Networked Storage supplier to join is noteworthy. The question is, will others follow? While it has joined the consortium, Qumulo has said little about how it intends to help the organisation. Nor which working groups it will join and have input into. One assumes that it will join the Storage Work Group. The advantage it will have is that if standards emerge from the discussions, Qumulo will be well placed to either influence or react to any decisions to ensure that it is first or early to market with any changes.
