ethernet vpn could be considered as the successor of vpls. it uses bgp for discovering involed pe routers. currently it have 3 different modes defined. the mpls mode introduced by juniper uses a plain mpls encapsulation in data plane and instead of mac address learning over the mpls core, uses client mac address advertisements within the bgp evpn afi. the pbb mode introduced by cisco uses pbb over mpls encapsulation in data plane, and uses client mac address learning at the edge pbb nodes which not necessarily match the mpls core boundary. so in best case no client mac addresses are handled at all at pe routers, but provider bridge mac addresses are advertised within the bgp evpn afi. the vxlan mode uses vxlan over raw ip as data plane and uses client mac address advertisements within the bgp evpn afi. all the mpls modes advertise a single label in the evpn afi because all modes use mac address advertisement and not learning. this is used as the inner service label and the outer, transport label is taken from the global table looking for an appropirate lsp for the remote pe router. here are the captures for mpls, pbb and vxlan. in case of evpn, no hierarchical version is introduced because the protocol is designed from the ground up to support large scale deployments thanks to bgp based mac advertisements instead of learning, with sites capable to multihomed to multiple pe routers. the bum traffic optionally could be carried by some p2mp technologies with the mpls involving modes.