Azure Sizing
Basics
SCEPman depends mainly on the CPU resources. Memory and disc are less important.
A SCEPman 2.5 instance in one Azure P0V3 App Service Plan (min. 195 ACUs) can serve around 2000 requests per minute under usual conditions. SCEPman 2.4 and earlier served about 400 requests per minute on one S1 App Service Plan (100 ACUs). Requests are
SCEP issuing requests and
OCSP requests.
Dependencies
The load for your SCEPman service has several dependencies and varies in the different environments. Important dependencies are:
Distribution of requests
Frequency of logins to network resources
Frequency of certificate requests/renewals
Especially the distribution of the requests has a high importance. If all clients do requests at the same time, your SCEPman instances get heavy load.
Please do not assign SCEP profiles to a large number of users/devices at once, since this may result in a request-peak at your SCEPman instances.
Recommendation
We recommend the following sizing in Azure Compute Units (ACU) for the Azure App Service Plans as a starting point:
< 2000 clients
~100 ACUs (e.g. 1 x S1)
2 x ~100 ACUs
(e.g. 2 x S1)
< 5000 clients
~200 ACUs
(e.g. 1 x P0V3)
2 x ~200 ACUs
(e.g. 2 x P0V3)
< 10.000 clients
~400 ACUs
(e.g. 1 x P1V3)
2 x ~400 ACUs (e.g. 2 x P1V3)
< 25.000 clients
~800 ACUs
(e.g. 1 x P2V3)
2 x ~800 ACUs (e.g. 2 x P2V3)
< 50.000 clients
~1600 ACUs
(e.g. 1 x P3V3)
2 x ~1600 ACUs
(e.g. 2 x P3V3)
< 100.000 clients
~3200 ACUs
(e.g. 2 x P3V3)
2 x ~3200 ACUs
(e.g. 2 x 2 x P3V3)
Azure Cost Prognosis
The main Azure costs will be for the Azure Apps Service Plan(s). You can derive your cost from the requirements in the table above and your Azure pricelist.
The additional Azure resources (Key Vault, Storage Account, Log Analytics, network resources for private endpoints) play a minor role in the cost. Depending on the use cases in your environment, you should expect an additional 5% to 25% on top of the App Service Plan cost for these additional Azure resources.
This cost projection is just a rule of thumb to help you estimate the cost of Azure. It can vary significantly in different environments.
Fine tuning
Every environment has its own load distribution over the day. In many environments the morning (start of work) generates a peak in terms of load at your SCEPman.
Manual Scale
You can adapt the computing power for your App Service to your individual daily load distribution with the Azure App Service Scale Out features. E.g. you could define 2 x S1 in the morning from 08:00-10:00 to cover the morning peak, while you reduce to 1 x S1 for the rest of the day.
Auto Scale
Alternatively you can use the Azure App Autoscaling feature to adapt to needed resources. Learn more about that in Autoscaling.
Manual vs. Auto Scale
If you are able to predict your load well (e.g. derived from load history), we recommend Manual Scale over Auto Scale, since Auto Scale has to behave lazy (hysteresis) to prevent flapping between scales.
Last updated