I just remembered the key technique to getting the communications working smoothly.
I do all my communication synchronously. That is as follows:
The master device sends a message to a node. The send command returns a success or fail, but fail does not mean the message was not received, it means that an ACK was not received. The master waits a timeout period for a response. If there is no response, a second attempt is made. If there is a response, the data with the response is processed. If not, the node is ignored.
After that transaction, the master repeats this process with the next node and for each node there after. A round-robin.
If a node does not get a request from the master within a certain interval, it reboots itself.
If the system power cycles, the nodes inundate the the master with DHCP requests. Because of so many requests, requests can be ignored or two nodes might receive the same network address (different from NodeID). Eventually, one node will get a valid network address and begin to communicate as expected. This reduces the network traffic and the probability of another node receiving a unique network address improves. Eventually, all nodes connect. "Eventually" can be many, many minutes.
The tree structure of the network means that some nodes send data through other nodes. In asynchronous communication, a node may be trying to send data while another node is sending data through it. I may not be correct in this, but I think that it is possible that this will corrupt one or both data packets. This is why I do the round-robin communication.
The master only has 5 (4?) usable addresses to receive data. Like computer DHCP, addresses can be reserved for specific NodeID's. For example, address 3 an be reserved for NodeID 31, then NodeID 22 can have a reserved address of 43, thus NodeID 22 will always send data through NodeID 31. (addresses are octal). The address 4444 is reserved; this is the address a node requesting an address uses.
To increase the chance of getting a valid address, I put a delay in the first DHCP request based upon the NodeID.
I put the NodeID in EEPROM address 0, as does MySensors. I have one program that sets the NodeID and then my program for all of my nodes is the same.
Nodes that sleep must have the flag set (in the node software) that indicates no data can be passed through it from another node.
All of these hoops are handled by the MySensors libraries, though I force the NodeID. This my impetus to use MySensors. I must wonder though, "How much overhead is there? Do I have enough memory for my code?" So far I haven't had a problem.
OSD
