Monitoring UDP drops causing pubsub to breakdown?

[chengguizi]

Hi eCAL team,

We have been using eCAL to send relatively high bandwidth data over LAN. There is an issue affecting smooth operation.

When we are sending a coupld of image streams concurrently, say the bandwidth is about 20MB/s, we notice the ecal Monitor at host machine will lose the publisher node on the edge machine.

I suspect this is due to the fact that, monitoring is transmitted using UDP, and natually it is unreliable, especially when the network is under load.

May question would be:

• What is the current behaviour, when eCAL subscriber node stop receiving ecal monitoring heartbeat, but UDP / TCP messages continuously to arrive?
• In other word, once the TCP / UDP sub node is intialised with first message received, will its transmission affected by failure of receiving eCAL monitoring heartbeat within the timeout duration?

Thanks!

[rex-schilasky]

Hi @chengguizi,

in general, it is not optimal to load the network so much that a large number of messages are lost. The eCAL registration (UDP) layer is sending heartbeat information with a default frequency of 1 Hz. See ecal.ini file

The monitor is collecting these information to monitor the whole eCAL system. This debug visualization has its own timeout, default setting is 5 sec. See ecal.ini

Publisher start sending payloads right after the first receive of a matching registration heartbeat and stop sending after a timeout that is configured in ecal.ini as well (default 60 sec), see

So, the publisher/subscriber may disappear in the monitoring visualization but transmitting payloads. But if the final registration timeout is reached they will stop sending until the get a new heartbeat message from their counterparts.

[chengguizi]

Thanks a lot for the insight! We will try to tune accordingly.