Azure Sizing

In many application areas, the use of cloud technologies is definitely a sensible and cost-effective alternative to traditional on-premise solutions.

The Questions:

Will a cloud database be able to keep up with an on-premise solution in terms of performance?
What costs are arising?

In the following section, we will use Microsoft Azure as an example.

The measurements were performed against an on-premise MSSQL reference environment

Detail Configuration OnPremise Environment

The following engines with WebTSMServices were used:

Service Maschine
OS	Windows 10 Enterprise x64 bit Version 10.0.17763 Build 17763
CPU	2 x Intel Xeon X5675 3.06 GHz
RAM	8 GB

Service Maschine

Windows 10 Enterprise x64 bit

Version 10.0.17763 Build 17763

CPU

2 x Intel Xeon X5675 3.06 GHz

RAM

8 GB

The following database was used:

	Database Server Properties
Server Version	Microsoft SQL Server 2016 (SP2-GDR) (KB4293802) - 13.0.5081.1 (X64)
OS	Windows Server 2012 R2 Standard 6.3 <X64> (Build 9600)
CPU	2 x Intel Xenon X5675 3.06 GHz
RAM	16 GB

DTU		S2 (50)	S3 (100)	S3 (100)	S4 (200)	S4 (200)	S6 (400)
App Service Plan		S2	S2	P1V2	P1V2	P3V2	P3V2
DTU Costs in EUR		65	125	125	250	250	500
App Service Plan Costs in EUR		75	75	125	130	500	500
Total Costs per Month in EUR		140	200	250	380	750	1.000
Delta to Referenz in %	Create 250 time series	614 %	493 %	462 %	672 %	442 %	490 %
	Write 250 time series for 1 month in 15 minutes interval	426 %	331 %	331 %	171 %	164 %	108 %
	Read 250 time series for 1 month in 15 minutes interval	264 %	203 %	97 %	59 %	27 %	25 %
	Read 250 time series for 1 month in 1 hour interval (aggregation)	115 %	114 %	97 %	90 %	62 %	79 %
	Aggregate 250 time series by 5 attributes	175 %	87 %	64, %	65 %	29 %	25 %
	Aggregate 250 time series in blocks of 50 by 6 attributes	126 %	101 %	63 %	55 %	32 %	31 %
	Delete 250 time series in 15 minutes interval	1164 %	960 %	955 %	464 %	473 %	227 %
Total score		244 %	184 %	146 %	100 %	72 %	60 %

The names of the scaling options offered are changing regulary. The above results are to be understood as a guideline.

Depending on the use case, more suitable and cost effective packages can be chosen, since, for example, the creation and deletion of time series is often not a time-critical process in daily operation.

Package Combinations

It makes perfect sense to combine packages for different core load times.

White to yellow packages are ideal for base load or for tasks with fewer IOs.
For performance comparable to a local installation, the green package is recommended, although yellow is also quite suitable due to the read performance, depending on the requirements.
For high performance applications the blue package is recommended.
The red package is only useful as a high-performance supplement in combination with lower packages due to the significantly higher price but comparatively low performance gain compared to the blue package.