For App Connectors, why shouldn't the customer pre-configure memory and CPU resources to accommodate a higher bandwidth capacity, like 1 Gbps or more?
In ZPA, App Connectors are designed to be lightweight, horizontally scalable components. Their effective throughput and concurrent-connection capacity are often constrained more by network stack limitations (such as ephemeral port exhaustion and per-process file descriptor limits) than by raw CPU or memory. As a result, simply over-provisioning vCPUs and RAM to ''hit'' a target like 1 Gbps on a single connector usually does not provide linear performance gains.
Zscaler design guidance emphasizes deploying multiple App Connectors and allowing ZPA to intelligently load-balance traffic across them. This delivers resiliency and scales capacity while staying within realistic limits of TCP/UDP ports and OS-level descriptors. Over-scaling a single connector can lead to diminishing returns and may even create harder-to-diagnose issues when port ranges or file descriptors are saturated.
Storage is not the main factor in App Connector performance, and the platform does not recommend a ''just throw more resources at it'' approach. For these reasons, the correct answer is that port exhaustion and file descriptors, rather than memory or CPU, are typically the true limiting factors for App Connectors.
===========
Gearldine
4 days ago