The key evidence is that at the median request load (p50), the service is already using all 8 of its maximum configured instances. This indicates that for at least half the time, the service is operating at full capacity. When all instances are busy, incoming requests are placed in a queue, leading to significant delays. This queuing effect is the direct cause of the p75 request time being 10 times longer than the p25 time. The server's low CPU (53%) and memory (44%) utilization confirms that the hardware is not the bottleneck; rather, the service's instance setting is too low to meet user demand.

A. There is no data provided, such as network latency or I/O wait times, to suggest network saturation. Low system usage on the database server also makes this an unlikely primary bottleneck.

C. The client operating system (Android) and server operating system (Linux) are not inherently a cause of performance issues and are irrelevant given the specific metrics provided.

---

1. Esri ArcGIS Enterprise Documentation. (2022). "Tune service instance settings." ArcGIS Server Administration.

Section: Configure service instance settings.

Content: The documentation states

"If you anticipate high traffic

you may want to increase the maximum number of instances. If all instances are busy processing requests

new requests are queued until an instance becomes available. If the request waits in the queue longer than the maximum wait time

it will time out." This directly explains the behavior observed in the question

where maxing out instances leads to long wait times (high p75).

2. Esri ArcGIS Monitor Documentation. (2022). "Key metrics." ArcGIS Monitor Help.

Section: Analysis > Charts and data.

Content: The documentation explains that percentiles (p25

p50

p75) are used to understand the distribution of response times. A large gap between percentiles

as seen in the question (p75 is 10x p25)

is a strong indicator of a bottleneck causing requests to queue. Correlating this with the "Instances In Use" metric hitting its maximum configured value confirms an instance-related bottleneck.

3. Esri. (2019). "ArcGIS Enterprise: Performance and Scalability Best Practices." Esri User Conference Technical Workshop.

Content: These workshops consistently identify insufficient service instances as a common performance bottleneck. They illustrate that when the number of running instances hits the maxInstances setting

response times increase dramatically while server CPU and memory may remain low. The solution presented is to increase the maxInstances value

provided the server has adequate hardware resources

which is the case in the scenario.