11 May
2016
11 May
'16
6:21 p.m.
On Tue, May 10, 2016 at 11:30:19PM +0200, Sipos Csaba wrote:
Maybe I am misunderstanding something, but I really
don't see what
is the reason behind handling the echo requests in the GGSN instead
of the gtp-u stack itself. Can someone please shed some light on
this one?
It's a typical design tradeoff for tunneling drivers: You keep
signaling side in userspace. We expect low rate of signaling messages
there, so this shouldn't harm performance. The kernel driver takes
care of G-PDU hotpath so we save context switches by handling
encapsulation/decapsulation without userspace intervention. On top of
this, bugs in the signaling don't crash your system, you can just
restart the daemon, and you can also easily update the userspace
signaling to get new features, while deploying a new kernel in a
datacenter is usually a more painful procedure.