Need larger sender queue in Akka Cluster

I recently upgraded to akka cluster 2.6 and noticed that in distributed pub sub outgoing messages starts dropping after queue size reaches 3k unsent messages. That is very different behavior from all previous versions; project started with Akka cluster 2.3 and even in 2.5 it still buffers enough messages to not drop on spikes.

I am not sure how to fix this, I read section about different types of mailboxes but mailbox is at receiving actor and messages are dropped in sender JVM before going over wire.

I found it, it is in the artery advanced settings section. I have to say that 3k default is way, way too low.

I think this is the first time we have heard about issues with that default being too low, so perhaps you are stressing remoting in a different way than most cluster users. Could there be some other root issue hiding there causing the messages to queue up to start with?