Profinet lets you wire almost anything — which is exactly why the plants with clean networks are the ones that decided not to.
The decisions
Line depth: daisy-chaining device to device is cheap and tidy, but every hop adds latency and a single point that takes everything behind it down. Practical ceiling for machine segments: single-digit hops for standard RT at moderate update times; deep lines also make port diagnostics ambiguous. Break long lines with a switch at logical machine areas.
Ring (MRP) where availability justifies it: one cable fault heals in the sub-second class (watchdogs must tolerate the reconfiguration time — configure them knowingly). Rings demand discipline: exactly one manager, documented ring ports, and no accidental extra loops — the “helpful” patch cable that bridges two switches is the classic broadcast-storm ignition.
Update-time budget: per-device update times multiply into bandwidth; drives at 2 ms, I/O at 8 ms, HMI traffic, plus reserve — the engineering tool computes load, and above ~50 % planned utilization you are borrowing from future extensions. IRT domains for motion get planned first, not squeezed in.
The port map is a deliverable: which device port connects where, printed in the panel door pocket and configured as reference topology — that is what makes neighborhood diagnostics and automatic device replacement actually work.
FAQ
Copper length limits? 100 m per segment, honestly measured including patch runs; beyond that, fiber — which also breaks ground-potential problems between buildings, a free bonus.
Unmanaged switches anywhere? In the office maybe; on the machine, Profinet-conformant switching only — LLDP neighborhood discovery is what your diagnostics stand on.
Zone Otomasyon delivers networks with port maps, load calculations and healing times measured. Network engineering.