The hyper scale public cloud providers such as AWS and Azure use the idea of instance or t-shirt sizes for their virtual machines. There is a massive range of alternatives depending on the CPU type, number of CPUs, RAM, the number of virtual disks that can be attached, whether they have GPU acceleration, etc.
Typically, these are all priced on a pay as you go or on-demand basis, where you are billed by the hour for the particular t-shirt size. Persistent storage is normally billed separately, unless you are using ephemeral VMs, where the storage associated with the VM is not persistent, and effectively disappears when the VM is turned off. In some cases, even with persistent storage, a virtual disk is often shown in a VM which is ephemeral and is used for caching or temporary space while the machine is running.
Customers can save money by turning off VMs at night and weekends, if it makes sense to do so. As well as the standard pricing models, other models have appeared recently:
- Reserved pricing – You agree to use the VMs 24/7 for a 1 or 3-year period. Large discounts are available, but you are paying for the VM regardless of whether you use it or not.
- Convertible reserved pricing – The same as above, but allows you to sell time slots for your reserved instance that you don’t need on a marketplace.
- Spot pricing – You bid for the maximum price you are prepared to pay for a VM. Depending on spare capacity, your VM will be powered on when your spot price is reached, and powered off again when your spot price is not high enough. Might be useful for batch processing requirements.
- Pay As You Go. Measure your actual usage of CPU, RAM and storage by the minute and then summarizing by the hour. Unlike the hyper scale public clouds, we do not mandate t-shirt sizes, customers can provision any size VM (up to some maximums), and can change the specifications whenever they like, even on the fly. A virtual machine may have up to 60 virtual disks, and up to 10 network interfaces, irrespective of the number of CPUs.
- Reservation pool. In this model, customers can reserve a ‘bucket of resources’; CPU, RAM, storage and internet, which is billed for per month at a reduced rate when compared to PAYG. A further benefit is that the resources are guaranteed to be available, and are not contented. When a reservation pool is full, it will not be possible to start any more VMs. Further discounts are available when committing to multiple year deals.
- The third model is to provide an amalgam of the two – a reservation pool with the ability to burst into extra PAYG resources as needed.
- With both of the last two models, it is possible to increase the reservation pool size through a call to the support team.