LACP on Viprion and Frame Distribution Hash

Hi,

 

I am rather not pro in this area so excuse me if my questions are not making sense.

 

I wonder what is relation between Frame Distribution Hash and VIPRION cluster distribution (end of article Overview of trunks on BIG-IP platforms.

 

I assume that flow is like that (Viprion uses vCMP, and multi blade vGuests):

 

• Peer switch receives packet destined to Viprion
• Peer is creating hash (based on its frame distribution algorithm)
• Hash is assigned to chosen interface in trunk
• All other packets with the same hash are send via the same interface
• Viprion receives packet and assigns it to TMM - this TMM can be running on vGuest VM on another blade than interface receiving packet - if I am not wrong this is possible?
• If TMM is on another blade then packet is send over backplane to VM on that balde
• According to article response packet will be send via interface on the blade where TMM actually processing packet is running, not via receiving interface

If above is correct is that not creating issue on the peer side? Will it not expect returning packet on the same trunk interface as send packet?

 

According to mentioned article:

 

When frames are transmitted on a trunk, they are distributed across the working member links. The distribution function ensures that the frames belonging to a particular conversation are neither mis-ordered nor duplicated at the receiving end.

 

What "conversation" means here?

 

• Only egress packets with the same hash value
• Both ingress and egress packets with same source IP:destination IP - I doubt, by not sure.

In the end I would like just to find logic for LACP:

 

On given side (peer or Viprion) LACP only assures that packets with given hash entering trunk will be send via the same interface.

 

Packets received from trunk (even if being response to given packet send into trunk) can arrive on completely different interface - LACP do not care to have one TCP connection or UDP flow to run via the same interface in both directions.

 

If above is true then choice of hashing algorithm can be different on both sides of trunk - Am I right?

 

Piotr