Building Automation Integration Services

What Sets the Poll Rate on an MS/TP Trunk

MS/TP is a token-passing bus over RS-485. One master holds the token, completes its transactions, and passes it to the next address in rotation, so every request on that trunk is serialized behind the token rather than running in parallel. Common line rates are 9600, 19200, 38400 and 76800 baud, and the whole trunk runs at the slowest rate every device on it can manage. One controller that cannot exceed 9600 sets the ceiling for the other thirty.

The Max_Master setting is the part most often left alone. Each master looks for its successor by polling addresses up to Max_Master, and the factory default of 127 means it keeps polling for masters that were never installed. On a trunk whose highest MAC address is 12, leaving 127 in place spends token time on addresses that will never answer, and the effect grows with the number of masters doing it. Setting Max_Master just above the highest address actually in use is a configuration change on each device, not a station setting, which is why it belongs in the survey findings rather than in commissioning.

Where a device supports change of value, subscribing is cheaper than polling: the station asks once and the device reports when the value moves outside the deadband. COV subscriptions carry a lifetime and have to be renewed, and support varies by device and by object, so a mapping usually ends up mixed. The rest is scheduling. A space temperature read every five minutes and a fan status read every fifteen seconds cost very different amounts of the same trunk, and deciding that per point is what keeps a large trunk usable.

Read Access and Write Access Are Separate Decisions

A supervisory layer will be disconnected at some point. A switch gets replaced, a controller is repowered, a link goes down during unrelated work. Safety interlocks, freeze protection and any sequence the equipment has to hold on its own therefore belong in the controller. If a sequence only works while the station is online, the sequence is in the wrong place. Read access and write access are scoped separately for that reason, rather than granted together because both were technically available.

On BACnet, a commandable object does not hold a single value. It holds a priority array of sixteen slots, and the value in effect is whatever sits in the highest-priority slot that is not null. A write at priority 8, the level normally used for manual operator commands, overrides one at 10 and is itself overridden by 1 and 2, which are reserved for life safety. Releasing a command writes null into that slot rather than a value, and the point falls to the next occupied slot below it.

When every slot is null, the object falls back to Relinquish_Default, which is the value the equipment runs on when nothing is commanding it. It is frequently left at whatever the manufacturer shipped, and it is worth reading before the first release rather than after. Modbus has none of this machinery. A holding register accepts whatever was written last, at whatever rate it was written, and nothing records the source. Two systems writing the same setpoint produce a value that alternates between them, which on a trend looks like an equipment fault rather than a configuration one.

For every point that will be commandable, the same short list gets settled and written down before the first command is issued:

• Read only or commandable, decided per point rather than per system, since one device usually has both
• The priority level the station writes at, and what else in the building already writes at or above it
• Whether the command is released after each use or held, and what releases it
• The Relinquish_Default the point falls back to, checked against what is safe with nothing commanding it
• For Modbus points, which single system owns the register, because the protocol will not arbitrate

The Network Path an Integration Depends On

BACnet/IP discovery uses broadcasts, Who-Is and I-Am, and broadcasts do not cross a router. Devices on another subnet therefore need a BBMD on their own subnet, or the station registers as a foreign device against a BBMD that already exists there. Two BBMDs on one subnet, or broadcast distribution table entries left behind by an earlier project, produce duplicated traffic and devices that show up twice.

Address translation is the other common surprise. When BACnet/IP traffic is forwarded, the originating device address travels inside the message, so NAT between the station and the equipment breaks discovery even though the packets arrive. Either the path is routed without translation, or the BBMD is configured with the translated address so the contents and the headers agree.

Firewall rules are where an otherwise finished integration stalls, because a blocked port and an unresponsive device look identical from the station. The default ports below are the ones to confirm are permitted between the station and each system, in the direction each one actually uses.

• BACnet/IP: UDP 47808 (0xBAC0), and 47809 upward where more than one BACnet network shares a subnet
• Modbus TCP: TCP 502, station to device
• MQTT: TCP 1883 plain or 8883 with TLS, outbound from the station to the broker
• SNMP: UDP 161 outbound for polling, UDP 162 inbound if traps are being received
• oBIX: over HTTP or HTTPS on whatever port the server publishes, commonly 443
• Niagara station to station: Fox on TCP 1911 or 4911 for the TLS version, with Workbench platform connections on TCP 3011 or 5011

What Happens to the Data After the Points Are Mapped

A point that reads correctly still needs a decision about history. Interval collection samples on a fixed period and produces predictable storage; change of value collection records only when the value moves past a deadband, which captures detail during transitions and nothing during steady state. A noisy analog with too small a deadband can generate more records than a one-minute interval would, and storage on an embedded controller is finite, which is why long retention usually means archiving to a Supervisor rather than keeping everything at the edge.

Timestamps deserve their own check, because there are two clocks involved: the device's and the station's. Where the station timestamps at poll time, the record says when the value was read, not when it changed, and the difference is the poll period. Energy and demand work is the least forgiving of this, since intervals get compared against the utility's own. The station gets its time from NTP, devices that keep their own clocks can be aligned with the BACnet time synchronization services, and the time zone and daylight saving settings have to match across the estate or a season change will introduce an hour of skew into a year of trends.

Alarms are where a new integration is most likely to annoy people on day one. Adding an alarm extension to every mapped point produces alarms from equipment that is simply off, from analogs sitting outside their limits during a shutdown, and from points that chatter around a threshold. Limits belong with a time delay so a brief excursion does not raise anything, and each alarm needs a class that decides who is notified and whether an acknowledgment is required. An alarm nobody acts on trains operators to ignore the ones that matter.