Qumulo says its Cloud Native Qumulo on AWS achieved over a terabyte per second throughput and more than one million IOPS using standard Network File System clients.
The company’s Core system software has been ported to Azure and AWS as Cloud Native Qumulo (CNQ). It offers both file and object protocols and Qumulo claims it has pricing “up to 80 percent lower than legacy cloud-based file solutions.” It also claims to be able to meet the performance and capacity needs of “virtually” any file-based application.
Steve Phillips, Qumulo’s head of product marketing and management, said: “For the first time, customers can use CNQ on AWS or Azure to dynamically scale performance or capacity, tailoring storage to meet changing workload demands from several gigabytes per second to now over a terabyte per second.”
He says CNQ has a cross-vertical appeal since customers in the healthcare and life sciences, media and entertainment, higher education, financial services, autonomous driving modeling, and energy verticals all rely on AI and high-performance computing to maintain a competitive edge.
CNQ has linear performance scaling and Qumulo says it delivers flash-class performance with an average cache hit ratio greater than 95 percent. From an AI processing point of view, CNQ “minimizes data load times and GPU idle cycles, reducing costs while speeding up time to training, tuning, and/or inference.”
Qumulo says standards-based CNQ has a collection of other benefits:
- Elastic scalability adjusts to changing workload demands, delivering performance when needed and throttling back to save costs when demand is reduced. “Unlike other cloud-based file systems,” CNQ operates without pre-provisioned capacity or rigid volumes, and clients pay for only the capacity and performance they use.
- Simplicity as customers can deploy CNQ in minutes and scale performance and capacity in seconds, freeing users to focus on business concerns while gaining faster access to their data.
- Global namespace (GNS) to enable low-latency access to data from any other location without complex replication or pre-staging operations, enabling customers to connect on-prem data to cloud-based AI and high-performance computing services.
- Cost savings as pricing is based on actual storage usage and performance delivered without requiring static pre-provisioned capacity.
Just as using AI and HPC is becoming a cross-industry feature, it is also becoming a cross-supplier feature. In the on-premises world, scale-out Qumulo is competing with parallel-access and GPUDirect-supporting filer suppliers such as DDN, Hammerspace, IBM’s Storage Scale, NetApp, Pure Storage, VAST Data, VDURA, and WEKA.
Dell is extending its PowerScale offering to add parallel file system access. Qumulo has GPUDirect support as a roadmap item. NetApp is extending its ONTAP file system with an ONTAP Data Platform for AI project.
It is happening in the object world as well. GPUDirect-type support is present with Cloudian and MinIO and coming to DDN Infinia and others. File and object storage suppliers are colliding as they try to offer the fastest, most power-efficient, and cost-effective way to feed unstructured data to AI training and inferencing GPU servers and HPC systems.
All the file and object storage suppliers are also emphasizing on-premises and public cloud hybridity, with global namespaces, common interfaces, and management.
Qumulo claimed a test said it was faster than WEKA in the Azure cloud in June, using an AI-related benchmark. It claimed at the time that this was both the industry’s fastest and most cost-effective cloud-native storage offering as its Azure run cost “~$400 list pricing” for a five-hour burst period. The SaaS PAYG (pay as you go) model pricing means metering stops when performance isn’t needed.
WEKA and Azure achieved 5GB/sec throughput in tests with the SMB protocol in February; a long way short of 1TB/sec.
We have no details about how CNQ passed the terabyte bandwidth milestone, and suspect that other cloud-native file storage vendors might achieve it if they throw enough nodes (cloud instances) into the pot. Certainly Qumulo has now set up a target for other vendors to reach.
Bootnote
CNQ is also available on AWS GovCloud, providing a secure, compliant option that meets government and regulatory standards.
