Hi @chuank
A lot of these problems are traceable back to firmware issues. There was a similar thread on a local cloud problem that turned out to be using up all the IP sockets in the core.
Can you try simplifying your firmware to the point where almost nothing can go wrong? You seem to have some variables in your firmware, try removing all the sensor code for these variables and just increment the values every time through loop or something similar. Boiling it down to the simplest test case will help a lot.
[EDIT]
One other thought--check your DHCP lease time and see if it corresponds to the time when the cores go zombie.