Model how least-connections balancing shifts new requests away from slow servers, and the arrival rate each member ends up with.
Because the algorithm always picks the member with the fewest live connections, connection counts converge to equal across the pool. Little's law then forces the arrival rate to be inversely proportional to hold time, so a server that takes three times as long receives a third of the new connections. Least connections is self-correcting for slow servers, which is exactly why it hides them — a member can be badly degraded and still look healthy because its connection count matches everyone else's.
Least Connections
least connections equalises held connections, so each server's arrival rate is inversely proportional to its hold time: rate_i = λ ÷ Σ(1 ÷ factor_j).
least connections equalises held connections, so each server's arrival rate is inversely proportional to its hold time: rate_i = λ ÷ Σ(1 ÷ factor_j). Because the algorithm always picks the member with the fewest live connections, connection counts converge to equal across the pool. Little's law then forces the arrival rate to be inversely proportional to hold time, so a server that takes three times as long receives a third of the new connections.
Least connections is self-correcting for slow servers, which is exactly why it hides them — a member can be badly degraded and still look healthy because its connection count matches everyone else's.
This calculator takes 5 inputs: Active connections across the pool, Servers in the pool, Servers responding slowly, How much longer a slow server holds a connection, New connections per second. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.