Let's start with a simple answer: performance is a unit of mass for data processing speed (in other words, for the number of calculations in a given period of time). Performance is measured at the system level (system performance) or determined for individual parts of the computer (for example, CPU performance or graphics processing unit (GPU) performance).

If that were the answer, we could stop now, but unfortunately, the situation in real life is not that easy.

As software architects, we are not, or are only marginally, interested in the system's performance. Instead, we are interested in the performance of our application (application performance) and it depends on several factors.

This includes the following:

• The code base of your application
• The performance of the underlying Azure services (components performance)

Regarding performances of the Azure services, the initial situation is more complicated for us because there are no uniform criteria for the performance terms.

Some examples are as follows:

Azure VMs and all services based on Azure VMs (Azure Cloud Service, Azure Container Service, Azure Service Fabric). Here, you set the performance level by selecting an Azure series. Which Azure series you choose depends on your individual workload and your specific requirements.

An Azure series (also known as VM type or VM instance) generally differs in the number of CPU cores, the amount of memory, and the maximum size of the data disk. Some Azure series or individual parts of it are also defined by special hardware.

Microsoft offers the following Azure series:

• A-series: The A-series supports the full feature scope of Azure VMs for a wide range of workloads.
• A-series (Basic): A light version of the A-series (only available for instances A0 –A4), designed for development and testing starter workloads.
• A-series (compute intensive): A high-end version of the A-series, designed for high-performance computing and data-intensive workloads (video encoding, for example). The series includes high-throughput network interfaces (RDMA) as an option.
• A-series version 2 (Av2-series): Series Av2 is the latest generation of the A-series and specifically designed for Azure DevTest Lab scenarios or for use as an Azure Service Fabric (a microservice architecture). With series Av2, the performance for data access is significantly higher, since only Solid State Drives (SSD) are used for data storage.
• B-series: B-series is intended for workloads that do not require full CPU performance continuously, such as web servers, or development and test environments.

Each instance of the BS-series provides a basic level (minimum value) for performance and then has the capability of extending to 100% CPU utilization of an Intel® Broadwell E5-2673 v4 2.3GHz or an Intel® Haswell 2.4 GHz E5-2673 v3 processor. As long as you use less than the basic performance with the VM instance, you acquire a credit balance. As soon as you are in credit, you can also use the performance that goes beyond the basic level.

All you have to do is to answer the question, which basic level best suits my needs?

• D-series and DS-series: D-series VMs are designed to run applications that demand higher compute power and temporary disk performance. To achieve this, a D-series VM provides a better processor performance and a higher memory-to-core ratio, and an SSD is used for the temporary disk.
• D and DS-series version 2 (Dv2-series and DSv2-series): Dv2-series and DSv2-series are the next generations of the D-series and DS-series. Dv2-series and DSv2-series are identical to the original series but have a 35% higher CPU performance based on the Intel Xeon® E5-2673 v3 (Haswell) processor.
• D-series and DS-series version 3 (Dv3 and Dsv3): Dv3-series and Dsv3-series are the latest generations of the D-series and DS-series. Dv3-series and Dsv3-series are based on the Intel® Broadwell E5-2673 v4 2.3GHz processor, or the Intel® Haswell 2.4 GHz E5-2673 v3 processor and introduce Hyper-Threading Technology and virtual CPUs (vCPUs) as key architectural elements. The new series provides also some of the first VMs to be running on a Windows Server 2016 host. Windows 2016 hosts enable nested virtualization and Hyper-V containers for these new instances. Nested virtualization allows you to run a Hyper-V server on an Azure VM.
• E-series and ES-series version 3 (Ev3-series and ESv3-series): The Ev3-series and ESv3-series are also part of the latest generation of the D-series and DS-series, and the new name for the high-memory instances D11-D14. For the Ev3-series and ESv3-series, the same description applies as for the Dv3-series and Dsv3-series.
• F-series and Fs-series: F-series and Fs-series offer a high CPU-to-RAM ratio, ideal for web servers, network appliances, batch processing, and medium-load application servers or real-time communication.
• FS-series version 2 (Fsv2-series): The Fsv2-series is the next generation of the Fs-series. The Fsv2-series are based on the fastest Intel® Xeon® Scalable processor, code-named Skylake (Intel® Xeon® Platinum 8168 processor), and featuring a base core frequency of 2.7 GHz and a maximum single-core turbo frequency of 3.7 GHz. Fsv2-series provides the Hyper-Threading Technology and the Intel® Advanced Vector Extensions (AVX) 512 instructions set, as key architectural elements.
• G-series and GS-series: The G-series and GS-series are a clone of the older D-series and DS-series, but the G-series instances have a memory twice as large, and temporary disks based on an SSD that are four times larger. With exceptional, high-performance VM sizes in the G range, you can easily handle business critical applications such as large relational database servers (SQL Server, MySQL, and so on) or large NoSQL databases (MongoDB, Cloudera Cassandra, and so on).
• H-series: The H-series is designed for high-performance computing and data-intensive workloads (for example, molecular modeling and flow dynamics). The series includes high-throughput network interfaces (RDMA) as an option.
• Ls-series: The Ls-series is a clone of the GS-series and built for low latency storage requirements greater the offerings of the GS-series.
• M Series: The M-series offers the highest number of CPUs (up to 128 vCPUs) and the largest memory (up to 2.0 TB) for a virtual machine on the Azure platform. This is ideal for extremely large databases or other high-end applications that require a high CPU count and large amounts of memory.
• N-series: The N-series is not defined by factors such as CPU or memory, but by the ability to use GPUs (the processors of the graphics card) as an additional power factor. Additional computing capacity is provided in the form of the GPU, whereby the GPU generally operates faster than the CPU in highly parallelizable program sequences (high data parallelism), and therefore, a simple way is sought to supplement the computing performance of CPUs by the computing performance of GPUs. The Compute Unified Device Architecture (CUDA) and/or Open Computing Language (OpenCL) technologies are used for this purpose. The use of the GPU performance is particularly suitable for computers and graphics with intensive workloads and supports you in scenarios such as high-end visualization, deep learning, and predictive analytics.

As you can see, there are some Azure series that have an S within the name. Those series VMs use SSDs and Azure Premium Storage as the primary storage type for data and operating systems.

Azure SQL Database: Again, you can choose between different options with the decisive difference that the underlying definition follows a different approach.

How is an Azure SQL Database performance level defined? A performance level is defined by the number of permitted Database Transaction Units (DTUs). The second criterion formerly used, the so-called transaction rate (a metric from the Azure SQL Database Benchmark (ASDB) to capture the maximum transaction throughput per hour, minute, or second) will no longer be used.