Really thank you.... I could fix the issue (on my local box still)..
The issue is because when the client is sending the message to the server, it will use a certain size of the memory.
Paging is using a lot of soft-cache and weak-references. Eventually the paged-message could go away, and at the ACK the size could be different than what it was originally. As a Consequence, the system was building up the difference up to the point it couldn't depage any longer due to the wrong counts on the size.
I still haven't committed as I had to add a lot of debugs that I don't want to commit. I should have it committed first hour on my morning.
thanks for the report again! I will update this thread as soon as I commit it. you will probably be able to get a patch easily for your test. We can talk about it tomorrow (on irc if you can).
I tested your changes on the 2_2_EAP branch and it looks good.