As we reported last fall, Oracle’s refresh of its Exadata database consolidation platform was a generation change with major architectural changes that turbocharge transaction processing and analytics. The Exadata X8M platform, introduced to on-premises customers last fall, is now becoming available in the Oracle Public Cloud, but with a major new twist. It is tapping the scale of the cloud to boost the capacity of Exadata for analytics to hold up to 25 PBytes of data.
Oracle Exadata customers have embraced the new platform. In Q1 results released last month, Oracle reported that X8M accounted for the majority of Exadata Database Machine sales, and that overall Exadata on-premises sales for the quarter grew 15%. That beat HPE, Dell, and IBM Power Servers which either recorded negligible growth or declines. The database cloud service (although, before now, not at the X8M generation) has been an effective draw for new business, with half of Exadata cloud customers being new to Oracle.
To recap, Exadata has always been positioned by Oracle as a high-performance, highly scaled self-contained data platform for Oracle Database where software and hardware are jointly optimized. As such, it used premium components such as InfiniBand for the backplane that, while a standard, was not widely adopted. The market for Exadata has been large enterprises with need for consolidating multiple databases with requirements for high transaction and analytic performance.
While still addressing the same target market, Exadata X8M marked a major architectural departure. The headline was adoption of Persistent Memory (PMEM), a new tier of storage that can deliver – if configured properly and in the right form factor – almost the performance of DRAM memory for transaction processing at a significantly lower price point.
But X8M also has several other key architectural changes as we pointed out last fall that are driven by embrace of industry formal and de facto standards. To recap, the first is a change in internal networking that directly impacts performance. X8M moves Exadata away from InfiniBand, a backplane that originally offered greater bandwidth, to 100 GbE Ethernet. That not only allows customers to use industry-standard networking with Exadata, but also opens it up to another high-performance protocol — Remote Direct Memory Access (RDMA) that the industry is now embracing over Ethernet with the RDMA over Converged Ethernet (RoCE) protocol. That allows much higher performance thanks to, as the name states, RDMA’s direct access from compute to memory, bypassing the entire OS, IO and network software stacks. That can be critical, especially for complex IOPS-intensive analytic problems. Next is the change of hypervisor, from Xen to KVM, which is a far more popular implementation in the cloud.
And it is in analytics where the new generation Exadata Cloud Service adds an extra benefit over on-premises X8M implementations. Traditionally, Exadata customers had a choice of capacity by rack sizing, starting with 1/8 rack and proceeding to ¼, ½, and full racks. For most customers, the sweet spot has been the ¼ – ½ rack sizes, which include two compute nodes and three storage servers in the ¼ rack. With X8M, those upper limits are increased to 32 compute and 64 storage servers, effectively boosting maximum storage to 25 PBytes through compression. That places Exadata X8M in data lake territory, a fact that Oracle exploits with its support of multiple data models, from relational to spatial, graph, and JSON.
The new Exadata X8M Cloud Service is available, both in the Oracle Public Cloud, and in hybrid/private cloud with Oracle Exadata Cloud@Customer and Dedicated Region Cloud@Customer. Given Exadata’s high-performance design point, Oracle points to its own benchmarks that show it outperforming cloud databases such as Oracle Database on Amazon RDS, and with Exadata Cloud@Customer, claims similar advantages over RDS on AWS Outposts. And given how Exadata Cloud Service scales compute and storage independently, Oracle is claiming that Exadata’s high performance and elasticity make better use of resources that should lower customers’ overall cloud costs because they can complete tasks faster with smaller configurations. To drive that point home, we’re waiting for the day when Oracle extends the Autonomous Database, which uses machine learning to further optimize compute, to the Exadata X8M generation of public cloud services.