ControlMaster vs. Independent SSH Connections: Selecting Based on Concurrency Load

Goal: Identify the concurrency level at which SSH connection multiplexing via ControlMaster ceases to provide lower average latency than establishing independent connections for each request.

Constraint: With ControlMaster enabled, the initial connection cost is amortized, but under high concurrency the shared multiplexed channel introduces contention for window size and encryption processing, causing latency to rise non‑linearly. The exact crossover point is not fixed in OpenSSH documentation and varies with server load, chosen cipher/MAC algorithms, and network round‑trip time, making a universal recommendation unavailable.

Questions: At what approximate number of concurrent SSH requests does the latency benefit of ControlMaster disappear compared to independent connections? How do server load and cipher selection shift this crossover point? What measurement approach yields a reliable estimate of this threshold for a specific deployment?