Users expect quick response times when they interact with ArcGIS services. A key contributing factor is the ArcSOC configuration (or service instances), which can be optimized in a few ways:
Optimizing your service instance configuration is not a one-time job. Usage patterns of your services can change over time, so balancing resources is an ongoing process. As an example, see test observations from Evaluating impact of adding mobile capabilities to a foundational network information management system.
In general, adequate ArcSOC processes are required to handle the load your services receive, and adequate server resources are required to support a given number of service instances. Because each busy ArcSOC requires an available vCPU, allocating too many ArcSOCs per vCPU can cause unacceptable wait times when vCPUs are busy. Further, too many ArcSOCs can also lead to excessive memory utilization. This is because each ArcSOC consumes memory related to your data and workflows. Conversely, allocating more server resources than necessary leads to unutilized capacity and unnecessary expenses.
Putting that all together, this means the ratio of running ArcSOCs to vCPU should be high enough that there are as many service instances as needed to support end users’ workflows, without exceeding acceptable resource utilization thresholds. A good general practice is that normal operation shouldn’t incur ArcSOC spin up wait time. For business-critical services that require predicable performance, consider setting the instance minimum and maximum to the same value.
The optimal ratio of ArcSOCs : vCPU will depend on your specific system and the work it performs. As a result, you can determine your system’s optimal ratio only through proper testing and observation practices. This test study looks at approaches to balance service instances with compute resources to help you get the best performance with your available resources.