In a bid to attract more small business and department-level customers to its high-end Exadata Cloud service, Oracle Corp. today launched what it calls an “intelligent data architecture” that delivers high-speed query performance across multiple cloud instances at a cost significantly lower than that of its flagship Exadata service.

Exadata Cloud is a high-powered database service used by 58% of the Fortune Global 100, according to Oracle. The cost of the enterprise-oriented service has limited its adoption by smaller companies. Oracle said the new architecture leverages economies of scale and elasticity to make Exadata more broadly accessible, with prices up to 95% lower than a dedicated service. It also won’t charge customers for input/output operations as most other cloud providers do.

Matt Kimball, vice president and principal analyst at Moor Insights & Strategy, called the low-priced offering “a big win for Oracle Cloud. All of the incredible features of Oracle’s database platform can now be easily consumed by companies of all sizes and all technical depth levels at an affordable cost of entry. This is a big deal,” he said.

Exadata Exascale uses a multi-tenant resource pool, hardware-based remote direct memory access and predictive preprocessing on storage servers to deliver query performance that Oracle claimed is more than 50 times faster than comparable services from other cloud vendors. RDMA allows one computer to access the memory of another directly without involving either one’s operating system.

Pooled resources

The data architecture is undergirded by virtualized database-optimized infrastructure on pools of shared computing and storage that enable elastic scaling and pay-per-use economics. The multitenant architecture provides elasticity and flexibility since customers can specify the number of cores and the amount of storage capacity they need. The Exascale control plane spreads every database across dozens of pooled storage servers. Physical server configuration is abstracted so users see only databases and virtual machine clusters.

“Customers don’t have to purchase database or storage servers,” said Kothanda Umamageswaran, senior vice president of Exadata and scale-out technologies at Oracle. “They can say they want 300 gigabytes of storage and four cores or 10 terabytes of storage and 200 cores. It’s elastic at initial provisioning and also after you purchase the system.”

Smart storage

Exadata Exascale is “a significant software change,” Umamageswaran said. “Previously, automatic storage management was used to distribute storage across databases. This new architecture has a common pool of Exadata compute and storage servers supporting thousands of tenants and millions of databases. Customers can come in as a tenant and specify just the cores and storage they need.”

Kimball said Oracle has achieved additional efficiencies “by taking databases and breaking them up into smaller extents that are optimally sized to deliver both performance and scale. This is a big deal as it enables [input/output] scale that allows what Oracle calls a loosely coupled architecture to work effectively.”

RDMA access is critical to improved performance because it forgoes the need to access data in storage through operating systems, databases, load balancers and other tiers. Databases can automatically fetch data using RDMA and pick it up from Exadata RDMA memory.

Oracle’s storage cloud uses x86 pre-processors to automatically tier data. The least-accessed, or cold, data stays on disk, warm data is loaded into a flash storage cache and the hottest data stays in memory.

”The storage cloud manages all the tiering transparently from the compute servers,” Umamageswaran said. “If the database, for example, sends a backup to the storage, we don’t automatically cache it; we’re smart about it and can send it directly to disk.”

The storage servers synchronize with the database to intelligently pre-load data based on query activity. “It actually knows the database rows and columns and what is the compression format,” he said. “All the database code is shared between the compute and the storage.”

Automated tiering

Tiering is “a function of what type of data you have and the workload profile,” Umamageswaran said. “It’s able to cache what the customer needs and ignore what the customer doesn’t need. For example, if you want the sales data from last week, it’s not going to send all of the sales data history back to the database.”

Oracle said the result is that Exadata Exascale combines the performance of dynamic random-access memory, the input/output performance of flash storage and the capacity of disks.

The architecture also enables virtual machine images to reside on intelligent shared storage, a function useful in a live migration of VMs. Space-efficient volume snapshotting and full clones of filesystems and databases for development and testing are available.

For organizations conducting large language model training, Oracle’s vector search can be transparently offloaded to Exascale storage and queries automatically parallelized across the storage cloud for better performance. Each storage server independently computes query matches, and the database merges the results, yielding what the company claimed is up to a 30-fold improvement in vector query performance.

Kimball said Exadata Exascale should appeal to departments within large organizations that couldn’t afford enterprise database services. When he worked in state government, budget-strapped agencies and working groups “would often go with a lighter database platform that lacked the capabilities and performance of Oracle,” he said. “This changes that equation entirely.”

That said, he added that Oracle isn’t a household name among small and mid-sized businesses. “If I were a marketer at Oracle, I would be building educational outreach campaigns to reach those market segments that wouldn’t even consider my technology at the outset because of perceived costs and complexity,” Kimball said.

Image: SiliconANGLE/Dall-E