Calculating Industrial Data Transfer: Protocol and Baud Rate

Calculating Industrial Data Transfer: Protocol and Baud Rate

## Why a "Standard Answer" Does Not Exist The original question "how do I calculate transferred data and pick a baud rate?" does not have a single closed-form solution. Network selection is a function of five coupled variables: - The volume of process data per cycle (the input and output image, in bytes) - The required update time (cycle, refresh, or scan period) - The distance between devices and the physical topology - The environmental conditions (EMC, cabling class, installed base, ingress protection) - The deterministic guarantee the application demands The methodology below converts these constraints into a numerical answer, then maps the result to a specific protocol profile. At the end, a worked example and a decision tree make the procedure reproducible on a real project. ## Foundational Definitions and the Slowest Link Principle Before any calculation, define the following parameters in a single table so the rest of the analysis is unambiguous.

Symbol	Definition	Unit
Nio	Total cyclic process I/O per scan (inputs + outputs)	bytes
Tcyc	Required cycle, refresh, or update time	ms
R	Net data rate of the chosen physical layer (baud or line rate)	bit/s
η	Protocol efficiency (overhead, inter-frame gap, retries, store-and-forward)	0 < η ≤ 1
tmin	Theoretical minimum transfer time on a single segment	s
Lmax	Maximum permitted cable length per segment	m
Nslaves	Number of active stations on the segment	count
Ttoken	Token rotation time (Profibus DP only)	ms

The fundamental inequality to satisfy is:

t_min = (N_io × 8) / (R × η) ≤ T_cyc

The right-hand side is given by the process; the left-hand side is determined by the chosen protocol stack and physical layer.

The Slowest Link Principle

A network is only as fast as its weakest segment. When the data path traverses multiple media (sensor cable to fieldbus segment, fieldbus to backbone switch, backbone to controller), the end-to-end update time is dominated by the segment with the lowest useful throughput, not the highest nominal bit rate. This is true for any IP-based industrial network and is the reason a 1 Gbit/s backbone does not speed up a 1.5 Mbit/s Profibus segment.

T_end-to-end ≥ max(t_{segment 1}, t_{segment 2}, …, t_{segment n})

Practical rule: measure or estimate the achieved throughput (not the nominal bit rate) for every segment. The sum of the segment latencies is the floor for the end-to-end cycle time. The Cisco learning material on link budgeting (Cisco: Calculating the time to transfer data across a link) describes the same principle for general IP networks and is directly applicable to industrial segments. ## Three-Step Sizing Methodology

Step 1 — Quantify the Process I/O Image

List every device that exchanges cyclic data with the controller and sum the input and output payload bytes per cycle. A typical sub-system table looks like this:

Device	Inputs (bytes)	Outputs (bytes)	Data exchange
SIMATIC ET 200S, 8 DI module (6ES7131-4BF00-0AA0)	1	0	Cyclic
SIMATIC ET 200S, 8 DO module (6ES7132-4BF00-0AA0)	0	1	Cyclic
SINAMICS G120 CU240E-2 with standard telegram 1	4	4	Cyclic
SITRANS P pressure transmitter (HART over Profibus)	5	1	Acyclic
SIMOCODE pro V PN motor protector	8	8	Cyclic
Total cyclic payload	18	14	—

The total cyclic I/O image for the controller is N_io = 18 + 14 = 32 bytes per cycle.

Step 2 — Specify the Required Update Time

The required update time is application-specific, not vendor-specific. The following table maps typical process classes to a sensible T_cyc.

Application class	Typical Tcyc	Notes
Slow process (tank level, temperature trending)	100 – 1000 ms	Modbus RTU, Modbus TCP, Profinet NRT, EtherNet/IP RPI 100 ms all acceptable.
Standard discrete I/O (conveyor, packaging)	5 – 20 ms	Profibus DP at 1.5 Mbit/s, Profinet RT 1 ms, EtherNet/IP RPI 2–5 ms.
High-speed discrete (capping, indexing, label applicator)	1 – 5 ms	Profinet IRT or EtherCAT at the field level; Profibus DP at 12 Mbit/s is borderline.
Servo / motion interpolated	0.5 – 2 ms	Profinet IRT, EtherCAT, SERCOS III only. Profibus DP and Modbus are excluded by definition.
Torque / current loop (drive-internal)	0.0625 – 0.25 ms	Handled inside the drive; the fieldbus only carries setpoint and feedback.

A motion application that calls for 1 ms cycle time has, by definition, eliminated Profibus DP, Modbus RTU, and most standard EtherNet/IP assemblies from the candidate set at the fieldbus level. Profinet IRT, EtherCAT, or SERCOS III remain.

Step 3 — Compute the Theoretical Minimum Transfer Time

For a single segment carrying only the cyclic image, the minimum transfer time (ignoring protocol overhead for the moment) is:

t_min = (N_io × 8) / R

Plug in the example from Step 1 with N_io = 32 bytes = 256 bits.

Candidate link	R (bit/s)	tmin (raw)
RS-485 at 9 600 bit/s	9 600	26.7 ms
RS-485 at 187 500 bit/s (Profibus DP)	187 500	1.37 ms
RS-485 at 1 500 000 bit/s (Profibus DP)	1 500 000	0.17 ms
Profibus DP at 12 000 000 bit/s	12 000 000	21.3 µs
Profinet IO at 100 Mbit/s	100 × 106	2.56 µs
EtherNet/IP at 100 Mbit/s	100 × 106	2.56 µs

The raw numbers look comfortable. The next section shows why the on-the-wire number is several times larger. ## Protocol Overhead and Efficiency Comparison The achievable efficiency η on a healthy, error-free segment is bounded by: - Frame preamble, start delimiter, header, and trailer - Inter-frame gap and token passing (for Profibus) - TCP/IP or UDP/IP encapsulation (for industrial Ethernet) - Switch store-and-forward latency (typically 5 – 10 µs per hop at 100 Mbit/s) - Application-layer acknowledgments and connection heartbeats Reference values, taken from official protocol documentation:

Protocol	Nominal R	Typical η (cyclic)	Comment
Profibus DP-V0	9.6 kbit/s – 12 Mbit/s	0.55 – 0.75	Token rotation adds latency; efficiency falls as slave count rises.
Profibus DP-V1	same	0.50 – 0.70	Acyclic read/write on top of cyclic traffic.
Profinet IO (RT, Class 1)	100 Mbit/s	0.30 – 0.60	Class 1 RT frames share the line with TCP/IP diagnostics.
Profinet IO (IRT, Class 3)	100 Mbit/s	0.60 – 0.85	Reserved time-slot transmission; jitter < 1 µs.
EtherNet/IP (CIP Class 1)	10/100 Mbit/s	0.25 – 0.50	Implied multicast; connection RPI governs actual update.
Modbus TCP	10/100 Mbit/s	0.20 – 0.40	Request/response per node; no hardware multicast.
Modbus RTU (RS-485)	9.6 – 115.2 kbit/s	0.70 – 0.85	High efficiency due to small frame, but polling is sequential.
EtherCAT	100 Mbit/s	0.85 – 0.95	"On-the-fly" processing in slave; highest η among industrial Ethernets.

Profibus DP efficiency curves are documented in the PROFIBUS System Description (PI Order No. 4.002, available from profinet.com), which is the canonical source for the 0.55 – 0.75 range and the relationship to the number of active stations and token holding time. ## Profibus DP: Baud Rate, Cycle Time, and Slave Limits For a Profibus DP segment, the achievable cycle time is:

T_cyc ≈ Σ (N_slave,i × 11 bit × T_bit) + N_slaves × T_slot + T_token

where: - 11 bits per byte (1 start + 8 data + 1 parity + 1 stop, with UART framing) - T_slot is the configured per-slave slot time (set in the GSD file; typical 10 – 100 T_bit) - T_token is the token rotation time, dominated by the configured T_tr (target rotation time, default 5 000 T_bit) - T_bit = 1 / R (bit duration) The Siemens PROFIBUS System Manual on the Siemens Industry Online Support portal (support.industry.siemens.com) provides worked numeric values for a 12 Mbit/s segment with 32 slaves, including the worst-case token rotation when one slave is absent.

Profibus DP Quick Sizing Table

Slave count	Recommended baud (≤ 1 km segment)	Max cycle at 1.5 Mbit/s	Max cycle at 12 Mbit/s
≤ 8	1.5 Mbit/s	< 1 ms typical	< 0.5 ms
8 – 16	1.5 Mbit/s	1 – 3 ms	< 1 ms
16 – 32	500 kbit/s – 1.5 Mbit/s	3 – 10 ms	1 – 3 ms
> 32	Consider Profinet IO	—	—

Beyond 32 active stations on a single segment, the token rotation penalty and the practical GSD parameter set make Profinet IO the better choice. Profibus DP also has a hard limit of 126 addresses, but practical segment stability degrades well before that.

Profibus DP Cable Length vs. Baud Rate

Baud rate	Max segment length (Type A cable, ≥ 0.64 mm)
9.6 kbit/s	1 200 m
19.2 kbit/s	1 200 m
45.45 kbit/s	1 200 m
93.75 kbit/s	1 200 m
187.5 kbit/s	1 000 m
500 kbit/s	400 m
1.5 Mbit/s	200 m
3 Mbit/s	100 m
6 Mbit/s	100 m
12 Mbit/s	100 m

Repeaters (e.g., 6ES7972-0AA02-0XA0) regenerate the segment and add another 100 m – 1 200 m depending on baud. Up to 9 repeaters in series are permitted, with a maximum of 127 stations on the overall bus. Source: Siemens PROFIBUS Network Manual. ## Profinet IO: Real-Time Class and Update Time Profinet IO defines several real-time classes, and the controller is configured with a "send clock" that determines the IO update time.

Real-time class	Minimum update	Jitter	Typical use
NRT (TCP/IP/UDP/IP)	100 ms	ms	Parameterization, diagnostics, web server
RT (Class 1)	1 ms	< 1 ms	Standard discrete I/O, variable-frequency drives
IRT (Class 3, "high performance")	250 µs	< 1 µs	High-end motion, isochronous synchronization

The update time is set in the device's PROFINET interface (e.g., in SIMATIC TIA Portal under Devices & Networks → Interface → Real-Time Settings). The achievable update time for a device is reported in the GSDML file, in the ApplicationLength and ReductionRatio fields. Always consult the GSDML before sizing. The Profinet commissioning guideline (PI document "PROFINET Installation Guide" Order No. 8.061, available from profinet.com) specifies the topology rules: maximum of 50 nodes per IRT island when using a cut-through switch with 1 µs latency class, cable length up to 100 m per segment (Cat 5e or higher, recommended Cat 6A for 1 Gbit/s).

Send Clock and Reduction Ratio

The effective update time of a Profinet device is:

T_update = T_sendclock × ReductionRatio

T_sendclock is global to the controller (default 1 ms). ReductionRatio is a power of 2 between 1 and 2 048, declared by the device. If the controller cannot satisfy the device's required ratio at the configured send clock, the device will fail to start up and report a configuration error in the PROFINET diagnosis. A common field failure: an S7-1500 is configured with a 0.5 ms send clock to satisfy a fast drive, but an older ET 200S station only supports a reduction ratio of 4 at 0.5 ms, meaning its update time is 2 ms. The 2 ms update does not fail the station, but it shows up as cyclic-jitter on the connected sensors. Always validate the worst-case ReductionRatio for the slowest device on the network before tightening the send clock. ## EtherNet/IP: RPI, Connection Limits, and CIP Motion EtherNet/IP does not use a single shared "cycle time" parameter. Instead, the controller opens a Class 1 connection with a Requested Packet Interval (RPI). The actual update is the maximum of (RPI, scanner poll period, connection heartbeat).

Scanner	Typical minimum RPI	Concurrent Class 1 connections
CompactLogix 5380 5069-L320	0.5 ms	180
ControlLogix 5580 1756-L85	0.2 ms	500+
Micro850 2080-LC50	5 ms	limited

EtherNet/IP performance is documented in the ODVA "CIP on EtherNet/IP Engineering Guide" (Volume 5, Edition 1.x), available from odva.org, which is the canonical reference for the DLR (Device Level Ring) latency and the implicit messaging "API = 4" connection timing. If a non-CIP-aware device or a third-party valve block must be polled, a per-message request/response exchange with typical 1 – 4 ms latency per node is the realistic bound. Multipoint polling is the de facto reason EtherNet/IP is rarely chosen for fast machine control with more than ~10 polled nodes per scanner. ## Modbus and Point-to-Point Serial: When They Still Win

Modbus TCP Frame Time and Polling Budget

Modbus does not have a publisher/subscriber model. The Master polls each slave sequentially. The total cycle time is the sum of the per-slave turnaround times. For a TCP frame on a 100 Mbit/s link with 8-byte payload (4 registers):

t_frame ≈ 12 (IFG) + 14 (Ethernet hdr) + 20 (IP) + 8 (TCP) + 12 (MBAP + PDU) + 4 (CRC) = 70 bytes ≈ 5.6 µs at 100 Mbit/s

The actual scan time per slave is dominated by the controller's task time (typically 2 – 10 ms per Modbus request/response in a typical PLC implementation) rather than the wire time. Always include the controller's request scheduling overhead in the cycle budget. The Modbus Organization's "Modbus TCP/IP Implementation Guide V1.0b" (modbus.org) is the official reference.

Modbus RTU on RS-485

For long-distance, low-speed multi-drop (e.g., a string of flow meters on a 1 km cable at 9 600 bit/s), Modbus RTU remains a valid choice. The wire time for a 16-byte request and 32-byte response at 9 600 bit/s is approximately:

t_rtu ≈ (16 + 32) × 11 / 9 600 = 55 ms

Add the silent inter-frame gap (3.5 character times, or 3.83 ms at 9 600 bit/s) and the per-slave turnaround. With 10 slaves polled sequentially, a 600 ms cycle is realistic, which is fine for a tank-farm application and unacceptable for a packaging machine.

Point-to-Point Serial (RS-232 / RS-485)

A direct serial link is still the correct choice when: - A single instrument is connected to a single controller with no need for multi-drop - Data rates are very low (a few bytes per second) at long distance - An existing installed base has spare COM ports and proven cabling A typical example: a weighing indicator (e.g., Mettler-Toledo IND780) on RS-485 at 9 600 bit/s, queried once per 100 ms with a 16-byte request and 32-byte response. The wire time is negligible compared to the controller's task scan; the protocol choice is driven by cost and availability, not by throughput. ## Worked Example — 64 Bytes, 10 ms Target A small machine has 64 bytes of cyclic I/O and must run at a 10 ms cycle. The candidate segment is Profibus DP, Profinet RT, or EtherNet/IP.

Step	Value
Required Tcyc	10 ms
Nio	64 bytes = 512 bits
Link R (Profibus DP, 1.5 Mbit/s)	1 500 000
Raw tmin	512 / 1 500 000 = 0.34 ms
Efficiency η (12 slaves, DP-V0)	0.65
Realistic t	0.34 / 0.65 = 0.52 ms
Headroom	10 / 0.52 = 19×
Link R (Profinet RT, 100 Mbit/s)	100 × 106
Raw tmin	5.12 µs
Efficiency η (RT, 12 devices)	0.45
Realistic t	5.12 / 0.45 = 11.4 µs
Headroom	10 000 / 11.4 = 877×
Link R (EtherNet/IP, 100 Mbit/s, RPI 5 ms)	100 × 106
RPI setting	5 ms (would meet 10 ms target)

The Profibus DP result is comfortable (19× headroom), but if 32 slaves must be added, the token rotation penalty and the configured slot time per slave consume most of that headroom. Recalculate before assuming the segment scales linearly. Profinet RT at this size is over-engineered; the limiting factor becomes the controller's OB1 (or OB61) cycle time, not the network. EtherNet/IP is acceptable if all devices are CIP-capable and a 5 ms RPI is the de facto project standard. ## Field-Proven Caveats and Selection Decision Tree

Caveats from Real Commissioning

1. Do not plan at 100 % utilization. Plan at 30 – 50 % to leave headroom for diagnostic expansion, additional acyclic read/write traffic, and retries during commissioning. The PROFIBUS System Description explicitly recommends a minimum 25 % margin. 2. EMC compliance does not change the throughput calculation. A borderline link budget that needs a baud rate reduction to recover errors is a sign of a cabling problem, not a sizing problem. Fix the cabling (shield bonding, equipotential bonding, distance to VFD cables) before re-sizing the protocol. Reference: IEC 61000-6-2 for industrial immunity, IEC 61000-6-4 for industrial emissions. 3. For Profinet, the configured send clock must be a multiple of the slowest device's reduction ratio. A "round" value like 1 ms may not be achievable for a given device even if the controller supports it. Verify each GSDML's ReductionRatio before locking the send clock. 4. For EtherNet/IP, the actual update time may exceed the RPI when the scanner task is overloaded, when the connection crosses a non-managed switch, or when QoS / DSCP is misconfigured on a routed path. 5. For Modbus TCP, the standard does not specify a maximum cycle time; the de facto limit is the controller's task period and the number of concurrent sockets. A 500-node Modbus TCP poll at 2 ms per request is not realistic; it would saturate the controller's task before the network. 6. Cross-vendor GSDs differ from vendor GSDs. A third-party Profibus slave with a non-certified GSD may have an inflated slot time that consumes more bus time than the equivalent Siemens device. Always read the GSD before specifying a slave count for the segment. 7. Watch the diagnostic slot. Most modern I/O modules reserve 2 – 6 bytes per station for diagnostic status. The cyclic image you see in the engineering tool is not the full on-the-wire payload.

Selection Decision Tree

1. Define T_cyc and N_io from the process specification, not from the controller's default values. 2. Compute raw t_min for each candidate physical layer using t_min = (N_io × 8) / R. 3. Apply realistic η using the table above; reject candidates whose realistic t exceeds T_cyc. 4. Add a 25 – 50 % margin for diagnostics, acyclic traffic, and retries. 5. Verify against the I/O device's GSD/GSDML/EDS file: does the device actually support the assumed update time at the assumed baud rate? Mismatches here are the most common field finding. 6. If multiple candidates remain, choose on installed base, EMC environment, and engineering tool preference, not on theoretical throughput. A "better" network that the maintenance team cannot diagnose at 02:00 is not a better network. ## Tools, Standards, and Official References

Siemens Documentation

- Siemens Industry Online Support: support.industry.siemens.com — search entries include the PROFINET System Description, PROFINET Commissioning Guideline, and the S7-1500 Communication Function Manual. - SIMATIC ET 200S Interface Module manual: support.industry.siemens.com (search "6ES7151-1BA02-0AB0"). - SINAMICS G120 Operating Instructions: support.industry.siemens.com (search "6SL3210-1PE-series").

PI (PROFIBUS & PROFINET International)

- PI home and document library: profinet.com — GSDML library, PROFINET Installation Guide (Order No. 8.061), PROFIBUS System Description (Order No. 4.002).

ODVA

- ODVA specification library: odva.org — CIP Networks Library, Volume 5 (EtherNet/IP), Volume 7 (CIP Motion).

Modbus Organization

- Modbus Organization: modbus.org — Modbus Application Protocol V1.1b3, Modbus TCP/IP Implementation Guide V1.0b.

Standards

- IEEE 802.3 — Ethernet standard, defining the nominal 10 / 100 Mbit/s and 1 Gbit/s physical layers used by Profinet, EtherNet/IP, and Modbus TCP. - IEC 61158 — Industrial communication networks, fieldbus specifications; covers Profibus, Profinet, EtherCAT, and others. - IEC 61784 — Industrial communication networks, profiles (PROFIBUS, PROFINET, EtherNet/IP, Modbus TCP). - IEC 61000-6-2 / IEC 61000-6-4 — EMC immunity and emission for industrial environments. ## FAQ