In the initial years of cloud computing, Pay As You Go (PAYG) or on-demand pricing was used as the compute cost model. The concept was revolutionary: Pay for the compute that you use and eliminate costs for unused resources. Pay for one virtual machine instance for two hours, and it might cost you $.50 instead of buying a physical virtualization host with a total cost of ownership (TCO) of $30,000. The cloud allowed organizations to streamline their IT cost models and become more nimble. Each year, more organizations and more workloads are migrated to the public cloud, to the tune of tens of billions of dollars. However, many have found that not keeping an eye on cloud consumption results in skyrocketing costs. So, reserved pricing was developed.
Reserved pricing allowed organizations to commit to an annual spend to gain a strong discount. But if you have elasticity in your model, this doesn’t work. Reserved instances assume full consumption of hours in a month. If you’re running a cloud compute instance on Amazon Web Services (AWS) or Azure 24/7, then you might consider agreeing to a reserved instance to get these discounts. However, let’s say you manage a university lab, and your lab compute instances in the cloud only run four hours a day. The reserved instance pricing doesn’t account for that and charges you for the entire 24-hour period. On-demand pricing only charges for the four hours each day and come to a less expensive daily total.
How Oracle changed the game
Public cloud with shared tenancy has since exploded, with many organizations throwing their hat in the ring, including Microsoft, AWS, and Oracle. Oracle initially created a cloud similar to first movers AWS and Azure, with on-demand pricing and one year and three reserved pricing. However, Oracle decided to rethink public cloud, which resulted in Oracle’s Generation 2 (Gen2) cloud, Oracle Cloud Infrastructure (OCI). This decision dramatically shifted how public cloud is managed and included single tier pricing, using one low price for infrastructure, platform, and software-as-a-service (IaaS, PaaS, and SaaS) services. No complicated tiered pricing, no conditional discounts. The pricing is predictable, simple, affordable, and always elastic—the way public cloud should be!
Take the university lab scenario, for example. If they use on-demand AWS pricing for their compute instances, such as a t3.xlarge Windows instance (burstable instance with 4 vCPU, 16 GB of memory, and a 40% bursting baseline) in the US Central region to run four hours a day for 1200 students. This configuration costs $24,964 per month, list pricing. By comparison, a Windows-based standard X9 shape in OCI with 2 OCPUs (cores), with 16 GB of RAM, and configured as burstable with a 50% baseline, costs $11,650.
Reserved instance pricing comparisons
Similarly, for AWS non-burstable instances, such as M6 class instances, where bursting isn’t an option, OCI pricing provides significant value. For example, an m6i.large Linux instance has an on-demand rate of $.096, while the OCI X9 Linux shape (or instance) with 8 GB of RAM has an effective hourly rate of $.066. Both these rates are elastic, and for a population of 1200 virtual instances, powered on 12 hours per day, the monthly average cost results in $42,048 for AWS and $28,908 for OCI. To get an EC2 rate low enough to match or beat the OCI rate, you’d need to use either one-year or three-year reserved instances. Each of these instances requires a commitment equal to its namesake. One-year reserved instance pricing for the m6i.large is 33.9% lower than the Ohio data center on-demand rate; three-year reserved instance pricing is 54.6% lower.
However, the reserved instances aren’t elastic. In committing to a one-year or three-year term, you’re agreeing to pay a per-month price, whether you’re running the instance 24/7 or not. The hourly rate is only reported for comparative purposes. In our 1200 virtual instance scenario, the cost for one-year reserved instances is $55,626 and $38,150 for three-year reserved instances.
The following image compares effective vCPU utilization for 100 AWS m6i.large instances to 100 X9 shapes during 24 hour. OCI charges per second for Compute shapes and is therefore highly elastic. AWS reserved instances charge per the term, not taking utilization into account, and charge for unused compute resources.
These pricing analyses compare AWS pricing in their least expensive region. All other regions of the world, such as Africa, Asia, Europe, and the Middle East, are much more expensive. OCI’s compute pricing is static in all 32 commercial regions. Incorporating other regions in the analysis only increases the cost differences. Let’s say that the 1200 virtual instances were split between the US Central (Ohio) and the Sao Paulo data centers. The OCI price of $28,908 USD remains the same, but the AWS on-demand cost increases to $54,531. The one-year and three-year costs jump to $60,606 and $49,371.
One argument against this line of thinking is that AWS and Azure cloud builds are not either elastic or non-elastic, but a mixture of both. “Use the on-demand where you need it and three-year RIs for your static workloads to reduce your overall cost.” While correct, this combination introduces complexity into the cost model and doesn’t consider storage or networking. Both options are 2–4 times more expensive with AWS and Azure over OCI. With the standard discounting and Oracle Support Rewards, you have a strongly positive total cost of ownership difference for OCI.
Conclusion
Many people lump Oracle into the legacy category when speaking of innovation and progress, but those people fail to see the innovations in Oracle Cloud Infrastructure. Cost effective all-NVME block storage, flexible Compute shapes, SSL off-loading, and static pricing (among others) are all innovations that contribute to the best price-performance results in the cloud industry.