Oracle already delivered what may be the fastest OLTP database machine ever built when it brought the Exadata X8M to market just over a year ago. However, today the company bested itself when it unveiled the new Oracle Exadata X9M. The X9M is a significant leap over what came before. This is a machine engineered to do only one thing: run Oracle Autonomous Database and Oracle Database faster and better than anything else on the market.
Wringing every ounce of performance and reliability from a database machine such as Oracle’s Exadata requires rethinking about system architecture from the ground up and with a deep understanding and appreciation of the intricacies of Exadata, and a deep knowledge of Oracle Database (which Oracle obviously does). Of course, you could run Oracle Database on a general-purpose server, but you would likely never achieve the peaks of performance offered by the X9M. The X9M, after all, is built by the same people who built the database that it’s designed to run. There is nobody better to take on that job than Oracle.
This may sound like hyperbolic praise, but once you look at what’s under the hood of the X9M, you’ll see that Oracle has earned the praise. The Oracle X9M is a stellar piece of engineering, bringing together compute and storage to deliver levels of throughput and reliability that deserve the superlatives that I’m throwing around.
One of this year’s most impactful technology events was Intel’s launch of its long-awaited Xeon Ice Lake server processors. Intel’s new parts, which it brands as its 3rd generation Xeon Scalable Processor, is a generational leap. In addition to the raw processor performance enabled by its new 10nm (true nm) process nodes, Ice Lake shines in powering the kind of I/O bandwidth required for a high-performance database like Autonomous Database on Oracle Exadata with its throughput-doubling PCIe 4.0 implementation.
It’s this raw processor performance, coupled with the faster I/O throughput enabled by PCIe 4.0 dual-port active-active 100Gbps network adapters, PCIe 4.0 flash cache, and 33% higher memory capacity (now 2TB per database server), to deliver some genuinely stellar raw performance numbers. The company says the Oracle X9M can provide 1TB/second of overall throughput and 27.6M SQL read IOPS per rack, which scales to even higher levels of performance as more racks are added.
Comparing the performance of the Oracle X9M against its X8M, the generational leap becomes apparent. Oracle tells us that the new machine costs the same as the old one but delivers 72% more IOPS in an OLTP environment, resulting in a 42% lower cost per IOP. The X9M also provides 33% more pooled resources for organizations that want to consolidate databases onto the smallest number of systems to drive up and resource utilization, and, of course, reduce TCO. Beyond OLTP, the X9M shines equally as well for analytics workloads. The X9M can deliver up to 87% higher throughput while lowering scan costs by 47% compared to the previous generation.
There is a lot to say about the raw performance of the Oracle X9M, and you can read about much of it on Oracle’s product page, but there’s more to the story. It isn’t all about new Intel Ice Lake processors and PCIe 4.0. Instead, database performance is as dependent on storage architecture nearly as much as it is on processor capabilities. It’s here that the X9M starts to look unlike anything we’ve seen before.
Storage & Cache
I’m primarily a storage analyst. I think a lot about how data can efficiently and safely move between the system components that rely on that data. So it’s almost an understatement to say that you’ve never seen anything quite like the X9M’s storage capabilities.
The storage architecture of the X9M includes aggressive use of shared persistent memory and RDMA-enabled interconnects coupled with dense high-performance NVMe Flash. There are also more traditional hard drives as part of the mix to handle the less active data. All of this is balanced to yield performance numbers that are close to unachievable in a more traditional three-tier storage/compute/networking architecture.
The Oracle X9M also has a unique cache architecture that places Intel Optane persistent memory (or “PMem”) cache in front of flash storage, delivering the first (and only) shared persistent memory cache optimized for a database engine. Oracle tells us that this allows for cache reads of less than 19 microseconds, less than nearly any dedicated storage array you can buy today and 10 times less latency than non-cached flash.
This cache architecture, along with the RoCE RDMA-enabled 100 Gb ethernet that I’ve already mentioned, allows the X9M to yield 10x better read latency and 8X faster log writes than earlier systems that didn’t use PMem and ROCE. The X8M already employed these two technologies, which makes the performance improvements and cost advantages of the X9M for both OLTP and analytics workloads all the more impressive.
Oracle has also increased the raw storage capacity of the X9M, increasing its overall capacity by 28%. A big part of that increase is swapping out 14TB drives for 18TB units.
Exadata X9M Cloud@Customer
Oracle makes the X9M available in what it calls Exadata Cloud@Customer X9M. The Cloud@Customer nomenclature follows Oracle’s tradition of naming systems based on what they do. Exadata Cloud@Customer brings Oracle’s high-performance Exadata Cloud Service inside customers’ data centers where they can meet data sovereignty or security requirements or simply use it as a first step towards running in the public cloud by allowing on-premises applications to have low-latency connectivity to a cloud database service. In this case, the word cloud is the critical one, because Exadata Cloud@Customer is owned and managed by Oracle and offers a flexible, consumption-based model running either Oracle Database or Autonomous Database on that infrastructure. It brings all the goodness of a cloud environment home to your data center with an Exadata as-a-service model. Our founder, Patrick Moorhead, calls Oracle’s Cloud@Customer the simplest hybrid cloud solutions in the industry and one that can be run off-line.
Exadata Cloud@Customer X9M brings all of the enhancements and performance capabilities already discussed but I believe does it with flexible and attractive cloud-like economics. I think it also could change the game when comparing Oracle’s on-prem cloud experience with what you’d find with a traditional cloud deployment or other vendor’s cloud on-premises offerings.
Oracle shows us, for example, that the Exadata Cloud@Customer X9M yields almost 50x to 100x better I/O latency in OLTP workloads than a similarly configured Azure SQL environment hosted on Azure. Oracle’s Cloud@Customer X9M also delivers up to 25x more throughput in analytics applications than what you can build in Azure. Of course, this isn’t all network latency. The Oracle X9M is designed for Oracle Database and Autonomous Database, while public cloud resources tend to be more general purpose.
More than 87% of the Fortune Global 100 depend on Oracle Exadata for business-critical workloads. This includes managing petabytes of data warehouses and delivering advanced data analytics. Exadata is also used to support complex business applications such as SAP, Oracle E-Business Suite, and Siebel. Where Oracle Exadata shines, however, is in servicing OLTP workloads such as financial trading and eCommerce.
The ability to have net incremental capabilities such as Smart Scan and Hybrid Columnar Compression are some of the key reasons that once customers try an Exadata with their Oracle Databases, they don’t want to go back to lesser alternatives.
All of this together delights Oracle’s customers. Wanthana Chotchaisathit, CIO of one of Oracle’s key customers, TISCO Bank Public Company Limited, had only praise for Oracle’s new offering.
Chotchaisathit highlighted the business value delivered by Oracle’s latest product, saying that the “Exadata Cloud@Customer X9M enables us to consolidate our business-critical Oracle databases to a single high-performance, high availability cloud platform in our own data center. This lowers our TCO and reduces the number of suppliers supporting our database infrastructure from five to one. As a result, our DBAs can now spend more time supporting our growth initiatives. Exadata Cloud@Customer X9M also helps us to better comply with banking data security regulations and policies which is very important to us.”
The Oracle X9M isn’t a machine that’s stingy with performance-impacting capabilities. You can start with a small, but still high-performance solution and grow it up to a full Exadata X9M rack which can contain a mixture of up to 1,216 database cores, 38TB of RAM, 3.8 petabytes of raw disk capacity, 920 TB of NVMe flash storage, and up to 27 TB of persistent memory. That’s a lot of horsepower, and Exadata’s flexible design allows you to configure it with the mix of compute and storage your workloads demand. The business-critical workloads that depend on Oracle Exadata can leverage every bit of the machine’s capabilities.
The availability of the new machine as both a stand-alone offering, as well as in a cloud on-premises Customer@Cloud model, makes it affordable to nearly every enterprise and Oracle Database customer.
Even mid-sized companies that have never considered an Exadata should do so now because the consumption-based pricing and that fact that Oracle manages the infrastructure – and the database with Autonomous Database – dramatically reduces the upfront and ongoing costs. Oracle isn’t announcing a cloud-hosted version of the X9M just yet, but I expect it will follow in the near-future. Until that happens, the X8M remains available in the Oracle public cloud.
I’ve just scratched the surface of what is new and different in the Oracle X9M, not even touching on the updated Exadata software and security features that are also being announced. I’ll be going deeper into all of this with a more comprehensive analysis in a few weeks.
If you run Exadata, you should have this machine. If you’re running Oracle Database on a DYI infrastructure, you should definitely investigate replacing it Exadata. If you’re just a fan of high-performance system design, then you should learn more about this machine. After all, Oracle may have just delivered the fastest database machine ever made. Again.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.