Azure Sizing

Basics
SCEPman depends mainly on the CPU resources. Memory and disc are less important.
A SCEPman instance in one Azure S1 App Service Plan can serve up to 400 requests per minute. 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:
1.Distribution of requests
2.Frequency of logins to network resources
3.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:
Amount of users/clients
Singular design
Redundant design
< 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 P1V2)
2 x 200 ACUs
(e.g. 2 x P1V2)
< 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)
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.

Amount of users/clients	Singular design	Redundant design
< 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 P1V2)	2 x 200 ACUs (e.g. 2 x P1V2)
< 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)

Last modified 3mo ago