At the most basic level, all storage performance boils down to hardware and in particular hard drives. Once upon a time, in the days of mechanical rotating platters, hard drive performance was measured in terms of rotational speed. A low-end notebook hard disk might spin at 4200 or 5400 rpm, while a high-performance server disk could spin up to 15,000 rpm. It's not just the disc's physical speed that counts but also the much faster cache memory built into the disc and the software smarts that manage the cache.

While spinning discs, that is, hard drives or HDDs are still in use and are suitable for high-volume low-performance storage, we mainly talk in terms of solid-state or SSD drives now. Instead of talking about rpm speeds, SSD drives are measured in random read/write operations per second. And like HDDs, SSDs come in various speed and performance configurations to meet different price points. These price points are reflected in the Azure managed disk options.

Ultra disk is the most performant disc option, followed by Premium SSD, then Standard SSD, while Standard HDD is still available for workloads where performance is unimportant. Not to state the obvious, but the higher the I/O throughput and the lower the latency, the more expensive the disc option.

Azure managed discs, also known as block level storage, can be thought of as physical hard drives or volumes when viewed from a virtual machine, even if that is not physically the case. Managed discs are unaware, and don't care what is stored on them, and are independent of virtual machines; that is, you can dynamically add more managed discs to a VM when the need arises.

When migrating a database from on-premise to the cloud, you will most probably want at least like-for-like disc performance. Because you can't specify the cloud's actual hardware, you need to base your disc choice on I/O operation metrics. There are three types of disc metrics that give a good indication of workload. There is the number of read and write operations carried out per second, also known as IOPS, input-output operations per second.

Throughput is the amount of data that is read or written from and to disk measured by bytes per second. Latency is the length of time taken for a read or write operation to commence once the request has been made. We can see straight away that the type of information or workload a database deals with places a different emphasis on these metrics. For example, when reading or writing large blob objects, like documents or even video, latency is far less critical than throughput in terms of the total operation's completion time.

On the other hand, when dealing with small amounts of data that are being written or read in real-time, latency becomes a more critical factor. As a rule of thumb, Ultra discs have the lowest latency, with a response time typically under one millisecond. Premium SSDs have the next lowest latency, usually in the low millisecond's range. Most production databases use either premium or ultra discs, or some combination of the two, but it usually comes down to a trade-off between performance and price.

You can use Performance Monitor to collect disk metrics on a Windows server. Once opened, select Performance Monitor under monitoring tools in the left-hand pane and then click the green plus button to add counters. Counter is another way of saying metric, as in counting the instances of an event occurring. After you have selected the server you wish to monitor, you are presented with a list of metric categories.

Find Logical Disk and expand it by clicking the small arrow on the list's right to reveal the different counters to track. IOPS can be tracked with disk reads, writes, and transfers per second, where transfers is the sum of reads and writes. Latency is monitored using average disk second read, write and transfer, while throughput is measured by bytes, read and written per second. It would be best if you also allowed an additional 20% headroom or margin for IOPS and throughput peak workloads.