|
Hello,
On Mon, 2 Apr 2001, Steve Gonczi wrote:
> Summary:
>
> In order to hand over an established TCp connection, both the source and the
> target
> of such hand-off has to be modified. (The current, "standard" TCP protocol
Hm, may be the source is not modified? After the connection
is moved then may be everything is similar to the direct routing
mechanism. client -> fron-end -> back-end -> client
> has no such capability).
> Given the above, there are many ways to do this.
> In the simplest scenario, there are 3 hosts involved:
> A,C, and D.
>
> A is sending tcp packets to C, and C decides to transfer the
> connection to D. In this scenario, all 3 need to be aware of
> the hand-off. (D needs to be told to fake up a connection,
> and start accepting packets from A, and A needs to be told to
> start sending the packets to D instead of C).
Hm, how A learns to send packets to D? I don't think this
is involved in the handoff mechanism.
> In a more real-life scenario, there are minimum 4 hosts involved.
> A,B,C, and D.
> A is a client, blissfully unaware of any handoff.
> B is a load balancer/relay that forwards and re-routes packets as needed.
> C and D are the actual hosts that negotiate the handoff.
If that is true, it looks very complex. As I understand,
the handoff mechanism delays the connection binding until it is
established and the request content is analyzed. Then this info
is used to schedule the connection to the "back-end" node. All
these games allow the Layer 7 to make better content-based
decisions for the scheduling. And we don't know the cost.
> /sG
Regards
--
Julian Anastasov <ja@xxxxxx>
|
