Static Arp Entry question. Cisco + NLB

I was working on a VMWare + NLB (Windows Network Load Balancing) project and had to put a static arp entry in the core L3 switch to get it to work. I understand that it is needed, but I do not understand why I didn't need to add it to the switch the host is connected.

L3 (Core)

---

uplink

---

L2 Switch

---

VMWare physical Host

I had to add the static arp to the L3 core with the MAC and IP of the virtual MAC and multicast virtual IP of the NLB cluster.

If I do a show mac-address-table on the L3 it doesn't show that that MAC is off of the uplink port. Also, if I do a show arp or show mac-address-table on the L2 switch I do not see the MAC address listed.

Any idea why this works? I thought layer 2 communication would need the MAC to show up on a physical port...

L2 devices do not ARP or respond to arp. ARP is a broadcast and L2 switches will do the normal thing (flood out of every port except incoming). ARP is needed on ethenet segments at L3 (Layer3 to Layer2 mapping).,

MAC address table at L2 is built from a device being connected to the switchport. switch adds the source mac address from a connected device once it sees the packet.
HTH.

My setup is technically this:

L3 (Core)

---

uplink

---

L3 Switch

---

VMWare physical Host

Your answer would still apply though because the second L3 switch isn't doing the routing because it's in a HSRP pair.

I've used NLB quite a bit i've never had to even think about arp. How many nics on this windows (virtual?) server are you connecting and when you configured NLB did you use unicast or multicast? I assume your trying to get NLB on a windows quest thats on your vmware host?