Oracle Database Appliance and CPU speed

Introduction

A complaint I heard from customers about ODA is the low core speed of the Intel Xeon processor embedded in the X8-2 servers: Intel(R) Xeon(R) Gold 5218 CPU @ 2.30GHz. 2.30GHz only? Because of its comfortable number of cores (16 per processor), the cruise speed of each core is limited. Is it a problem compared to a home made server with less cores?

Why clock speed is important?

As you may now, the faster a core is running, the less it takes time to complete a task. Single core clock speed is still an important parameter for Oracle databases. Software architecture of Oracle is brilliant: automatic parallelism can dramatically reduce the time needed for some statements to complete, but the vast majority of them will be processed on a single thread. Regarding Standard Edition 2, parallelism does not exist on this version, thus each statement is limited to a single thread.

Is ODA X8-2 processor really limited to 2.3GHz?

Don’t be affraid by this low CPU speed, this is actually the lowest speed the cores are guaranteed to operate. Speed of the cores can be increased by the system, depending on various parameters, and fastest speed is 3.9GHz for this kind of CPU, which is nearly twice the base frequency. This Xeon processor, as most of its predecessors, features Turbo boost technology, a kind of intelligent automatic overclocking.

Turbo boost technology?

As far as I know, all the Xeon family has Turbo boost technology. If you need more MHz than normal from time to time, you CPU speed can greatly increase to something like 180% of its nominal speed, which is quite amazing. But why this speed is not the default speed of the cores? Simply because running all the cores at full speed has a thermal impact on the CPU itself, and the complete system. As a consequence, heating can exceed cooling capacity and damage hardware. To manage speed and thermal efficiency, Intel’s processor dynamically distributes Turbo bins, which are basically slices of MHz increase. For each CPU model, a defined number of Turbo bins is available and will be given to the cores. The rule is that each core will receive the same Turbo bins numbers at the same time. What’s most interesting on ODA is that it’s related to enabled cores: the less cores are enabled on the CPU, the more Turbo bins are available for each single core.

Turbo bins and limited number cores

With limited number of cores, the heating of your CPU will be quite low in normal condition, and still low under heavy load because the heatsink and the fans are sized for using all the cores. As a result, most of the time, the Turbo bins will be allocated to your cores, and if you’re lucky, you’ll be running at full throttle, meaning that, for example, instead of a 16 cores CPU running at 2.3GHz, you’ll have a 4 cores CPU running at 3.9GHz. Quite nice isn’t it?

With Enterprise Edition

One of the main feature of ODA is the ability to configure the number of cores you need, and only pay the license for these enabled cores. Most of the customers are only using a few cores, and that’s nice for single threaded performance. You can expect full speed at least for 2 and 4 enabled cores.

What about Standard Edition 2?

With Standard Edition 2, you don’t need to decrease the cores on your server because your license is related to the socket, and not the cores. But nothing prevent you from decreasing the core numbers. There should be a limit where less but faster cores will benefit to all of your databases. If you only have a few databases on you ODA (let’s say less than 10 on a X8-2M), there is no question about decreasing the number of cores: it will most probably bring you more performance. If you have much more databases, the overall perfomance will probably be better with all the cores running at lower speed.

And when using old software/hardware?

Turbo boost was also available on X7-2, but old software releases (18.x) do not seem to let the cores go faster than normal speed. Maybe it’s due to the Linux version: the jump from Linux 6 to Linux 7 starting from earlier versions of 19.x has probably something to do with that. Patching to 19.x is highly recommended on X7 for a new reason: better performance.

Conclusion

If you’re using Standard Edition 2, don’t hesitate to decrease the number of enabled cores on your ODA, it will probably bring you nice speed bump. If you’re using Enterprise Edition and don’t plan to use all the cores on your ODA, you will benefit from very fast cores and leverage at best your licenses. Take this with a grain of salt, as it will depends on the environment, both physical and logical, and as these conclusions came from a quite limited number of systems. Definitely, with its fast NVMe disks and these Xeon CPUs, ODA is the perfect choice for most of us.