A pooled ArcGIS Server service has a specific count of related process instances needed for performing asynchronous processing of requests in parallel.
The minimum and maximum number of instances count have big impact on the overall performance of a WebOffice 10.8 SP2 project (project load/start up time as well as for server round trip time).
Depending on the estimated count of concurrent users (users accessing WebOffice 10.8 SP2 in the same period of time, more precise in the same minute), the minimum and maximum number of instances must be increased accordingly.
Recommended minimal setting for at maximum five concurrent users is
•Minimum number of instances: 1
•Maximum number of instances: 3
The number of instances can be specified either in the service editor during the publishing process or later on in the ArcGIS Server Manager.
Service Editor - minimum and maximum number of instances
ArcGIS Server Manager - minimum and maximum number of instances
Note: For rarely used map services (e.g. which only are used in a single WebOffice 10.8 SP2 in a single map view) it might be an option to set the minimum number of instances: 0. Although the first attempt to use this service may take longer, you can save resources for heavily demanded services.
Note: The calculation of number of instances below is a rough "thumb rule" like calculation and may not provide good results in specific configurations/scenarios. E.g. if only dynamic map services are used (no caching) then the probability for concurrent requests will increase – as a logical consequence the number of instances will need to be increased.
For each five additional concurrent users (i.e. within peak load time hour of the day, within time interval of one (1) minute, five additional users access WebOffice 10.8 SP2), the overall instance count should be increased by at least 1.
E.g. if there is an ArcGIS Server service used in a WebOffice 10.8 SP2 project and an estimated count of 15 concurrent users access the WebOffice 10.8 SP2 project using the ArcGIS Server service, then the recommended minimum instance count would be 3:
Base count + (overall concurrent users per minute - 5) / 5
1 + (15 – 5) / 5 = 1 + 2 = 3
Using the maximum number of instances parameter it is possible to better cope with load peaks. Thus, if you have very high load peaks but a small base load, then you may, e.g. increase the maximum number of instances by one or two counts.
It is not recommended to increase or decrease the number by more than one instance at a time and always have a representative time span to reanalyze the changes. This way side effects (e.g. impact on other services) can be minimized.