Attached is a diagram, would it be fine if glbp is run across the devices R1 & R2 at location B.R1 & R2 are 2 different service providers link which end up on the L3 switch.Would this work for outwards loadbalancing.All hosts are behind the switch?
Kindly suggest the problems with this setup also ..as it seems glbp has issues with address resolution.
Victor, you're tecnically correct, since GLBP replies to host ARPs with its virtual gateway MACs, just one host MAC would not preclude load balancing.
For instance, if our single host was dealing with GLBP configured for round-robin load balancing and it ARPed before sending each IP packet, we would effectively have achieved load balancing (similar to CEF's packet-by-packet).
Practically, though, since hosts don't usually ARP for every IP packet, we're not going to see concurrent load balancing when there is only one host MAC.
Also from a practical level, there's reasons why IP doesn't normally ARP per packet. Further, although IP itself doesn't require flow packet sequencing, it's something we try to maintain.
So, for practical reason, there's a problem when GLBP only "sees" one host MAC although you're correct that it's more to do with GLBP only seeing ARPs from the one host. Because, in a round-robin GLBP configuration, when the host's ARP time-outs, GLBP will move the host to the other gateway, but during the ARP cache time period, only one GLBP gateway will be used.
Since GLBP does respond to ARPs, all flows from one host will go to the same gateway. Further, ARP times can run into minutes. Given these facts, many routing protocols might effectively load-balance better than GLBP. Given the L3 switch, in this instance, this is one reason why I recommended routing vs. using GLBP.