For example, as hosts are added to an Ethernet LAN, the broadcast and multicast traffic
load typically increases due to hosts issuing periodic ARP requests or searching for
local network resources (this is protocol dependent, of course).
The LANE and MPOA architectures in the 1990s attempted to address Ethernet
scaling, but used a very complex design including dedicated servers for forwarding
broadcast, multicast, and unknown packets. This experiment was not a great success;
hence VPLS adopted a simplified architecture where the PE devices themselves replicate
those packets.
One specific issue when connecting Ethernet switches to a VPLS instance is that the
VPLS learns all MAC addresses for the switched network. This constrains VPLS scaling
as the PE devices have to learn all active MAC addresses in the switched network.
For this reason, many service providers offering VPLS service either limit the number
of MAC addresses per attachment circuit, effectively restricting customers to attaching
routers, or charge customers for blocks of MAC addresses.
MPLS 447
H-VPLS The flat topology of VPLS (where every PE has a pseudowire to every other
PE) causes two major scaling challenges:
?– The ingress PE has to replicate broadcast, multicast, and unknown frames to all
other PEs in the VPLS.
Pages:
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010