Andrew May, Cloud Solutions Architect, writes…
Virtual Machines (VMs) and Managed Disks in Azure come in pre-defined sizes, each having baseline performance capabilities for Input/output Operations per Second (IOPS) and throughput, with larger resource sizes having greater performance. More recently, these resources have gained the ability to burst beyond their baseline.
The size of a Managed Disk defines the storage capacity in GB as well as the IOPS and throughput. If more storage space is needed, the Managed Disk needs to be changed to a larger size. However, if the storage space is fine, but more performance is needed, a bigger Managed Disk size still needs to be chosen, even if the extra storage capacity isn’t used.
The ability to burst beyond this limit, creates potentially huge savings for a business as they do not need to deploy larger resources in order enable the higher IOPS. If your daily peak performance time is less than the burst duration, you can use bursting VMs or disks as a cost-effective solution. You can build your VM and disk combination, so the bursting limits match the required peak performance, and the baseline limits match the average performance.
Disk-level bursting allows the VM to boot faster and handle spiky workloads where increased performance is only needed occasionally, all while keeping the Managed Disk size, and therefore costs, as small as possible.
Depending on the Managed Disk size, two types of bursting are used: Credit-based bursting for smaller Disks or On-demand bursting for larger Disks. On-demand bursting is currently in Preview, so I’ll focus on Credit-based. While a disk is performing below its baseline, any unused IOPS or throughput is credited to a burst bucket. When the disk needs to perform above its baseline, those credits are used up. The bucket always starts off full when the disk is deployed, it’s enabled by default and the best bit – Credit-based bursting is free!
In summary, bursting lets you deploy smaller VMs and Disks that are aligned your average requirements, rather than your maximum requirements, while still being able to handle short-term demand for increased performance and this ultimately saves you money.