LAN SDN technology hard core insider 10 cloud converged matchmaker evpn

In yesterday's theme 《 From layer 2 switching to layer 3 routing 》 in , We already know Neutron and VMWare NSX In the virtual network model ,VXLAN How the forwarding plane works . that , Packet forwarding depends on MAC and FIB Where does the information come from ？

Early Neutron In the version , Adopted Openflow agreement , take MAC,ARP and FIB It is distributed to OVS in .Openflow Appearance , It is a major change in the network industry , but Openflow There are also some fatal defects .

One side ,Openflow The mechanism instead of flooding learning leads to the overload of the control node .Openflow VXLAN In the network ,VTEP Between BUM Messages will be triggered Openflow The control node answers or distributes the flow table , When the network is large , It is a heavy burden on the control node .

On the other hand , If the control node stops unexpectedly ,Openflow The stream table will not be refreshed , The new virtual machine can't go online . This is unacceptable in a production environment .

because Openflow These inherent defects of design , People need new control plane protocols , by VXLAN The network synchronizes the MAC and FIB Information .

Some people noticed VXLAN and VPLS The similarities of .VPLS There are two control planes ,CISCO Dominant Martini The way is to use LDP As the control plane （RFC4762), and Juniper Dominant Kompella The way is to use MP-BGP As the control plane （RFC4761）.VPLS The main problem is the need for all BUM Data packets are flooded to achieve MAC Address learning , And waste valuable backbone network bandwidth . To solve this problem , People have come up with EVPN(RFC 7432), adopt BGP Protocol to exchange MAC/FIB Information .

We found out , If you use VXLAN Instead of VPLS, Also is to EVPN Act of agreement VXLAN Control plane of , It can also be realized in the data center network VXLAN Layer 2 and layer 3 forwarding , Here's the picture ：

EVPN Of Route Type2, It can be realized in various VTEP Synchronization between VM Of MAC、IP、 On the second floor VNI( The subnet you are in )、 Three layers VNI( symmetry IRB The way to do it VXLAN When routing , The public rotor network used VNI). EVPN Another way to declare , It's called Route Type5：

Route Type5 and Route Type2 The difference between , The former can be in the form of subnet prefix , To others in the network VTEP Declare IP Address segment , The latter is declared discrete in the form of host routing VM（ host ）. therefore ,Route Type5 It is very suitable for connecting network boundary routers VTEP, To the other VTEP Announce outgoing route .

that , In the picture RR What role is it ？ Let's recall 《BGP Design and implementation 》 Mentioned in , stay AS domain , all iBGP A full connection should be established . such , If there is n individual VTEP, In total, maintenance is required n(n-1)/2 A connection . introduce RR, Then you can put iBGP The number of connections from O(n2) Reduced to n.

With EVPN As VXLAN Control plane of , We found that , The two problems mentioned above have been solved .

EVPN Instead of flood-and-learn The mechanism of ,VM When going online or migrating ,VTEP As long as the announcement is updated ;

EVPN After configuration , various VTEP Will remember the relevant configuration , The control node can only add tenants / Subnet time , Will modify EVPN To configure , Avoid the risk of single point of failure of the control node .

EVPN It can be said to be the matchmaker of cloud network integration ！