What is uptime?
Cloud uptime is the length of time that a service provided by a cloud platform is available to users. In other words, uptime is the amount of time that you can access your computing workloads in the cloud. The key to managing uptime is understanding your workload type, and matching it to the best-fitting cloud instance type.Batch Workloads
Batch workloads are interruptable and non-interruptable workloads. They run hourly, daily, monthly or any other discrete time for a specific duration. Batch workloads may include end of month reporting for accounting systems, banking account daily and month end updates, protein folding, graphics rendering, machine learning & training runs. Most cloud platform providers have VMs that can be created from “leftover” or “residue” hardware resources at a huge discount to regular on-demand VM prices. These are called spot instances on AWS and Microsoft Azure, preemptible VMs on Google and Oracle clouds, or transient VMs on IBM Cloud. You can secure up to 90% cost savings when you use spot instances compared to on-demand VMs. For example, preemptible VMs on Oracle Cloud are always 50% off your on-demand VM rates. However, in most cases your VMs can be taken away from you at ANY time if someone outbids you, usually with 1-2 minutes notice. So they’re not suitable for mission critical workloads that cannot be interrupted!Always-On Workloads
Always-on workloads are exactly what they sound like–workloads that cannot be interrupted and must run 24 hours a day, 365 days a year. These may include customer facing websites, database servers for worldwide ERP systems, worldwide file servers, and payment gateways. Servers that need to be “always-on” can benefit from reserved instance discounting. Reserved instances when you commit to continuous usage for a long period of time. If you select a longer term reserved instance, you can realize a deeper discount. For example, you can get an average discount of 30% from on-demand VMs from a 1 year reserved instance, versus a 70% discount for a 3 year reserved instance. However, you forgo the ability to upgrade to new releases of VM models during the term commitment. Since mission critical workloads are “always-on,” there is no opportunity to manage its uptime manually, so reserved instances are your best bet for cloud cost savings.Sometimes-On Workloads
The last main workload type is a sometimes-on workload. Sometimes-on workloads operate in certain time windows but are rarely needed outside of those time windows. They can include print servers, corporate intranets, and point of sale servers. On-demand instances are the best option for these kinds of workloads. Additional cloud savings can be attained by properly managing the on and off times for your VMs that are running in the cloud. This differs substantially from on-premises environments where the underlying server is kept on all the time. Certain workloads are appropriate for this kind of uptime management, especially ones that are tied to business operating hours. For example, corporate servers may only be necessary between 9am-5pm. Applications that run intermittently can help save costs in the cloud because you don’t pay for resources when they are turned off. For instance, an analytics application that runs only 20% of the time costs $4000 per month in AWS, rather than $41,000 on-premises. This represents a savings of 90%! The bottom line is: shut your VMs off when not in use to maximize savings.
Summary
It’s important to split up your workloads and map them to the types of VMs on the cloud platform to reduce your cloud costs. Batch workloads generally fit best with spot instances, always-on workloads fit best with reserved instances, and sometimes-on workloads fit best with on-demand instances. That said, these guidelines are not one-size-fits-all and every company is different, so it is critical to look at your unique computing needs and capabilities before making cloud instance decisions. After you have tagged your workloads as batch, always-on, or sometimes-on, you will need to model the pricing and sizes of the VMs for your newly categorized workloads to optimize for costs overall.
