@@ -14,6 +14,7 @@ metadata:
14
14
spec :
15
15
ipv4Namespace : default # Limits which subnets can the VPC use to guarantee non-overlapping IPv4 ranges
16
16
vlanNamespace : default # Limits which Vlan Ids can the VPC use to guarantee non-overlapping VLANs
17
+ mode : " " # Empty string is the default meaning l2vni, other option is l3vni
17
18
18
19
defaultIsolated : true # Sets default behavior for the current VPC subnets to be isolated
19
20
defaultRestricted : true # Sets default behavior for the current VPC subnets to be restricted
@@ -197,3 +198,61 @@ spec:
197
198
- from: 1000
198
199
to: 2999
199
200
` ` `
201
+
202
+ # # Mode
203
+
204
+ VPCs can operate in two modes : L2VNI and L3VNI. L2VNI is the default mode of
205
+ operation and represents the conventional functionality. L3VNI is designed
206
+ for switches that lack the hardware support for L2VNI.
207
+
208
+ # ## L2VNI Mode
209
+
210
+ This is the conventional multi-tenant network virtualization mode. It is the
211
+ default option for VPCs.
212
+
213
+
214
+ # ## L3VNI Mode
215
+
216
+ In L3VNI mode, the switches are configured to exclusively route unicast traffic.
217
+ This enables multi-tenancy inside of a fabric, even with switches of mixed
218
+ capabilities. The [DS5000](../reference/profiles.md#celestica-ds5000) is an
219
+ L3-only leaf and VPCs attached to this switch must be in L3VNI mode. VPCs in
220
+ L3VNI mode are not able to use switches configured for ESLAG.
221
+
222
+ Without broadcast traffic, each end host needs to have a full /32 address for
223
+ its address (e.g., `10.10.0.5/32`, not `10.10.0.5/24`). The host also
224
+ needs to emit traffic containing its IP-to-MAC mapping before the network will be
225
+ able to route traffic to it, as there is no MAC learning.
226
+
227
+ The DHCP server included with the Fabric has been updated to support L3VNI
228
+ mode. When a VPC is using the included DHCP server and is in L3VNI mode,
229
+ the DHCP server will send a DHCP lease with a short duration, so that the DHCP client will immediately request a new
230
+ lease. The DHCP renewal traffic allows the network to detect the host and redistribute the route via BGP.
231
+ Subsequent lease requests will use the configured lease duration.
232
+
233
+ If a user elects to use their own DHCP server or statically assign IP addresses, it
234
+ is recommended that the user set the following `sysctl` values on the end hosts :
235
+
236
+ ` ` ` console
237
+ net.ipv4.conf.default.arp_notify=1
238
+ net.ipv4.conf.default.arp_announce=1
239
+ ` ` `
240
+
241
+ # ### Example Route Output
242
+
243
+ If the fabric DHCP server is enabled and serving a default route :
244
+
245
+ ` ` ` console
246
+ user@server ~$ ip route
247
+ default via 10.10.0.1 dev enp2s1.1000 proto dhcp src 10.10.0.4 metric 1024
248
+ 10.10.0.1 dev enp2s1.1000 proto dhcp scope link src 10.10.0.4 metric 1024 # Route for VPC subnet gateway
249
+ ` ` `
250
+ If the fabric DHCP server is enabled and not serving a default route :
251
+
252
+ ` ` ` console
253
+ user@server ~$ ip route
254
+ 10.10.0.1/24 via 10.10.0.1 dev enp2s1.1000 proto dhcp src 10.10.0.4 metric 1024 # Route for VPC subnet gateway
255
+ 10.10.0.1 dev enp2s1.1000 proto dhcp scope link src 10.10.0.4 metric 1024
256
+ ` ` `
257
+
258
+
0 commit comments