Communication services · 2 August 2016
Polling and Grouping
Fetching points one by one is too slow for big BMS installations. Grouping several points into one message speeds polling up, and here is how Modbus, BACnet and others do it.
Polling mechanism described here and here does not have throughput big enough to ensure high update rate for big BMS installations. The next enhancement we will introduce is point grouping.
The idea is simple: instead of getting points one-by-one, let’s request and receive multiple points at once. Naturally, every message will become bigger, but not proportionally bigger. Receiving two points instead of one might make a message only about 15% longer. That’s because apart from point values, messages contain a header, length, device address and cyclic redundancy check, whose size is unchanged.
| Silence | Address | Function | Byte count | Data | CRC | Silence | Total | |
| 1 point | 3.5 | 1 | 1 | 1 | 2 | 2 | 3.5 | 14 bytes |
| 2 points | 3.5 | 1 | 1 | 1 | 4 | 2 | 3.5 | 16 bytes |
This shows the structure of two Modbus RTU responses: the first with one point, the second with two. Each silence is 3.5 character times, counted here as bytes. The full response grows from 14 to 16 bytes (+14%) while the payload doubles (+100%). That’s good scalability.
As always, there are extra requirements to implement grouping:
- Device CPU should be fast enough and have bigger network buffer
- Longer messages are more error-prone – better network quality is needed
- Software should be smart enough to group points effectively
Let’s review how grouping is implemented in a few popular protocols.
Modbus
Modbus always uses point groups – even a single point request is a group with just one member. There are a few limitations, though:
- Only one type of points could be grouped together. E.g. input registers with input registers, not with coils
- Groups include sequential points only. E.g. holding registers from 10 till 13, not 10, 12, 19
- Maximum size of a group varies between devices, thus software can’t group points fully automatically and should rely on engineer’s input.
BACnet
BACnet has probably the most advanced mechanisms to poll group of points. Read Property Multiple service allows to request various properties of various objects. As BACnet devices are much better standardized than Modbus and also have services to specify their memory limitations, software can fully automatically generate polling requests in the most optimal way.
BACnet also defined a Read Property Conditional service – as its name implies, it returns values only if defined criteria are met. It could reduce traffic even more, but few devices ever implemented it, and the service has since been deprecated.
GENIbus
Grundfos GENIbus telegram can include multiple Application Protocol Data Units (APDU). Each APDU contains multiple points of a single class (type). So in practice, it is easy to request many points of different types in one request.
LonWorks
Nope, no grouping here. And actually, it is not necessary, because polling is a second class citizen in LonWorks. This protocol has much better mechanism, called binding, which is essentially a subscription.
Sauter novaNet
Sauter novaNet addressable memory is quite big. Each hardware and software point in addition to real-time value stores various descriptions, units, types, timestamps and other properties. To deal with it, novaNet offers very advanced grouping capabilities. Points could be requested:
- horizontally – multiple sequential properties in a single point area
- vertically – single property in multiple sequential point areas
- randomly – any combination of points and its properties