which kind of MPLS deployment is the next-hop-self all keyword used on a BGP neighbor command?

Since the core and aggregation parts of the network are integrated and end-to-end LSPs are
provided, the Unified MPLS solution is also referred to as “Seamless MPLS.”
New technologies or protocols are not used here, only MPLS, Label Distribution Protocol (LDP),
IGP, and BGP. Since you do not want to distribute the loopback prefixes of the PE routers from
one part of the network into another part, you need to carry the prefixes in BGP. The Internal
Border Gateway Protocol (iBGP) is used in one network, so the next hop address of the prefixes is
the loopback prefixes of the PE routers, which is not known by the IGP in the other parts of the
network. This means that the next hop address cannot be used to recurse to an IGP prefix. The
trick is to make the ABR routers Route Reflectors (RR) and set the next hop to self, even for the
reflected iBGP prefixes. In order for this to work, a new knob is needed.
Only the RRs need newer software to support this architecture. Since the RRs advertise the BGP
prefixes with the next hop set to themselves, they assign a local MPLS label to the BGP prefixes.
This means that in the data plane, the packets forwarded on these end-to-end LSPs have an extra
MPLS label in the label stack. The RRs are in the forwarding path.
There are two possible scenarios:
In both scenarios, the ABR sets the next hop to self for the prefixes advertised (reflected by BGP)
by the ABR from the aggregation part of the network into the core part. If this is not done, the ABR
needs to redistribute the loopback prefixes of the PEs from the aggregation IGP into the core IGP.
If this is done, there is no scalability.
In order to set the next hop to self for reflected iBGP routes, you must configure the neighbor
x.x.x.x next-hop-self all command.
Reference. http://www.cisco.com/c/en/us/support/docs/