As stated before, these applications are receiving traffic from SCTP socket (~ 3.6 Mb/s), moreover they are connected to UDP socket.

More debugging of the problem shows the below results:

1- The applications are not hanging but they are being inactive waiting for traffic from the SCTP sockets.
2- Each time the I/O activities run the UDP socket dies.
3- Each time the I/O activities run The SCTP socket is still alive and we can see it in /proc/net/sctp/assocs however, no traffic is received anymore from this socket and that's why the applications seem to be dying.

Hence, it seems that these I/O activities are affecting the kernel sockets and not my applications.

Is there anyway to optimize the kernel parameters regarding this behavior?

Whether it is a constant load or due to heavy bursts, it's clear the IO is the bottleneck, not the CPU. Could be the driver buffering or the apps.
As the OP wrote the apps, he should know how they work, including how they get the data from the driver.
In either case, I guess a ramdisk could help channel the load/bursts away form the disk towards the RAM, but it could require the apps to be adapted.