Quanta Cloud Technology QuantaMesh QNOS5, Configuration Manual

Page: 248 / 256

248
Overall, the system has a maximum allowed limit of 4096 static host MAC-to-VTEP bindings. At any point in
time, the sum of all tenants static host MAC-to-VTEP mappings must be less than or equal to the system
limit. Once this limit is reached, configuring new MAC-to-VTEP bindings for any tenant results in failure and
a log message is generated.
The operator may optionally configure host MAC-to-access port entries as well. The maximum allowed
static host MAC-to-interface bindings (or local tenant system MAC entries) per interface (physical or Port-
channel) is 24. Once this limit is reached, configuring new MAC-to-interface bindings for any tenant results
in failure and a log message is generated.
9.6.2.8. ECMP
A tunnel may have multiple next hops when the underlay has multiple next hops to the tunnel's remote
endpoint. Many data center designs make heavy use of ECMP. To get good traffic distribution within the
underlay, it is important that encapsulated packets hash well.
VXLAN encapsulation includes a UDP header. Switches can include the source and destination UDP port in
ECMP hash computations. The hardware offers an option for the source VTEP to set the source UDP port to a
variable value (hash based on incoming packet Ethernet header) to ensure good ECMP hashing. VXLAN
enables this option in hardware by default.
Note: At VXLAN initiation, payload fields are used for hashing at the egress and also to generate the
entropy into the UDP source port which becomes part of VXLAN tunnel information. This UDP source port can
be used by transit switches for hashing purposes.
9.6.2.9. MTU
VXLAN encapsulation adds 50 bytes of overhead. This additional overhead can cause an encapsulated packet
to exceed the MTU of the outgoing port. The gateway does no IP fragmentation while tunneling a packet and
is by default configured to set DF=1 in the outer IPv4 header. If an encapsulated packet exceeds the L2 MTU
of the outgoing port, the hardware drops it. To avoid this problem, operators must ensure that the L2 MTU
on gateway ports to the underlay and underlay network be configured at least 50 bytes larger (for VXLAN)
than the MTU on ports on the access side.
The hardware may also enforce an IP MTU. In most cases, network-side ports will be configured as port-
based routing interfaces. The IP MTU of these routing interfaces will automatically be adjusted to match the
L2 MTU. Therefore, if the administrator adjusts the L2 MTU as described above, the hardware should not
drop packets because of an IP MTU limitation. If, however, network-side ports are VLAN routing interfaces,
the administrator will need to also increase the IP MTU on each network-side routing interface.
9.6.2.10. TTL and DSCP/TOS
By default, the switch is configured to behave as follows:

The TTL in the outer IPv4 header during tunnel encapsulation is set to 255.