> From: Heng Qi <hen...@linux.alibaba.com> > Sent: Thursday, March 16, 2023 9:39 AM > A rough description of this scene is as follows:
> Client1 sends packets to client2, and client2 sends packets to client1 > respectively. > This is called two directions, and they are in the same flow. > The packets in the two directions may reach the processing host through > different tunnels. In this scenario, we need to hash these packets to the same > queue for the same cpu to process, so inner symmetric hashing is required. > > > client1 client2 > | | > | __________ | > +----->| tunnels |<-------+ > |---------| > | | > | | > | | > v v > +-----------------+ > | processing host | > +-----------------+ > I understood now. The key piece is the processing host is different from clients 1 and 2, I was missing that. The key is there are two tunnels and processing data on the inner header for the different tunnels is your primary requirement. Hence inner hashing helps to serialize them.
