22.05.2017 Views

nx.os.and.cisco.nexus.switching.2nd.edition.1587143046

Nexus Switching 2nd Edition

Nexus Switching 2nd Edition

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Figure 9-102 shows the VXLAN packet format.<br />

Figure 9-102. VXLAN Packet Format<br />

A virtual vmknic network adapter is created on the ESX h<strong>os</strong>t <strong>and</strong> assigned to the VEM. This<br />

enables the VEM to encapsulate the original L2 frame with the VXLAN identifier. Today,<br />

VXLAN is supported only on the Nexus 1000v; if there is a requirement for external<br />

communication of the VXLAN segment the following solutions are available:<br />

• Deploy a multihomed virtual machine with two interfaces. One interface would reside<br />

in the VXLAN network, <strong>and</strong> the second interface would reside in the VLAN network.<br />

• Deploy a VXLAN gateway; the Virtual ASA (vASA 1000v) or Cloud Services Router<br />

(CSR 1000v) can be deployed as a VXLAN gateway.<br />

The following requirements are needed to deploy VXLAN:<br />

• The MTU needs to be increased by 50 bytes to enable the VLAN MAC-in-IP<br />

encapsulation.<br />

• Recommend using LACP port-channels for the PNICs.<br />

• The upstream physical switches from the VEMs need to have IGMP querier-enabled.<br />

• Proxy-ARP needs to be enabled on the Layer3 Interfaces (SVI) upstream routers for<br />

the port-profiles that are capability vxlan-enabled.<br />

• Multicast routing needs to be enabled on the upstream switches or routers if the VEMs<br />

are separated by routers <strong>and</strong> require VXLAN connectivity.<br />

• Enable 5 Tuple Hashing for optimal load balancing for LACP; the encapsulation will<br />

generate a source UDP port based on a hash of inner packet 5-tuples.<br />

The default setting for the VM vNICs (vEthernet) MTU is 1500 bytes. This is also true for<br />

the uplink pNICS (Physical Ethernet) uplink interfaces. With the default setting of 1500 byes<br />

on the edge MTU, a minimum of 1550 bytes is needed to allow for the VXLAN<br />

encapsulation. On the Nexus 1000v, the additional MTU can be enabled on the Ethernet<br />

uplink profile assigned to the physical ESX h<strong>os</strong>ts; nothing else needs to be configured on the<br />

VM or h<strong>os</strong>t side. The physical upstream switches that provide the connections need to have<br />

jumbo frames enabled as well. Another option is to set the edge MTU on the VM vNICs to<br />

1450 bytes, which would be 50 bytes less than the default 1500 bytes.<br />

For proxy-arp, it is enabled by default on IOS-based devices. For NX-OS platforms, proxyarp<br />

is disabled by default <strong>and</strong> needs to be enabled.<br />

For the Multicast, multicast queries need to come into the vmk NIC for the VXLAN transport<br />

VLAN. This requires that IGMP snooping querier be configured for the transport VLAN.<br />

Without the querier enabled or configured, the multicast will not work <strong>and</strong> there will not be<br />

any remote IP interfaces learned for VXLAN operation.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!