1. Requirements for DRNI on layer 3 gateway devices
802.1 AX-REV, IEEE 802 plenary
Lu Huang

2. Agenda
Use case
Requirement
Proposal

3. Use cases Use case 1: Mobile backhaul DRNI
In LTE mobile backaul, one optional solution is deploying routers and layer 2 network(PTN). For the interconnection between routers and PTN, DRNI is an appropriate mechanism
Router is the Layer 3 gateway of eNodeBs. In this case, DRNI is used on layer 3 gateway devices
Furthermore, In the best case convergence time should be less than 200ms, in the worst case it should be less than 500ms
DRNI
Use case 2: Layer 2 network in cloud computation IDC
In cloud computation IDC, we use layer 2 network to meet the requirement of cloud computation. We want to use DRNI instead of spanning tree (Of course, SPB or Trill are also optional, maybe in the future)
Core switch is the layer 3 gateway of servers. In this case, DRNI is used on layer 3 gateway devices
Furthermore, convergence time should be less than 500ms
DRNI
VM migration
Searching
Datebase access
DRNI
We want to use DRNI on layer 3 gateway devices
We require fast convergence in DRNI for protection

4. Agenda
Use case
Requirement
Proposal

5. Requirement(1) Unique MAC for DRNI
Layer 3 network
For hosts there should be only one gateway, so GW1 and GW2 should share the same MAC and IP address for LA1 (the distributed link aggregation group)
When hosts request GW’s ARP, GW1 and GW2 should answer the same ARP ( GW_IP, unique MAC)
When GW1 and GW2 request hosts’ ARP, they should use the same sender ARP info ( GW_IP, unique MAC)
Because GW1 and GW2 will use LA1’s MAC as their ARP MAC, so DRNI should make sure that nodes in the same DAS share an unique MAC which could be used by upper layer protocol (e.g. ARP)
We can config the same IP on LA1 of GW1 and GW2. It doesn’t depend on DRNI protocol
GW1
GW2
LA1
LA2
L2SW1
L2SW2
Layer 2 network
L2SW
Host
ARP Request
ARP Reply

6. Requirement(2) L3 entries Backup mechanism
To obtain fast protection switch, GW1 and GW2 should share the same L3 entries (e.g. ARP entries). When GW1 or link between GW1 and L2SW1 break down, GW2 could forward all flows immediately
Real-time backup
When GW1 receives ARP request or ARP reply packet from host, it should duplicate one and send to GW2
When GW2 receives ARP request or ARP reply packet from host, it should duplicate one and send to GW1
GW1 and GW2 could limit the duplication based on the ETH_TYPE of packets ( for ARP, it’s 0x0806)
This mechanism could not used in the below scenario
Batch backup
When GW1 reboots, batch backup(with GW2) can help GW1 recover forwarding all flows as soon as possible. Otherwise GW1 wil take even a few seconds to relearn ARP entries and introduce a long interruption
How to fulfill batch backup is under research ( through ICCP, LACP or ARP )
Of course, if we define a protocol for batch backup, it also could be used for real-time backup
Layer 3 network
GW1
GW2
L2SW1
L2SW2
Layer 2 network
Host
ARP Request or Replay
Layer 2 network
Layer 3 network
GW1
GW2
L2SW1
L2SW2
Host
Batch backup message

7. Agenda
Use case
Requirement
Proposal

8. Proposal
When using DRNI on layer 3 gateway devices, DRNI should support the following features
Nodes in the same DAS share an unique MAC for LA group
A node could duplicate some certain packets to the other nodes in the same DAS based on ETH_TYPE of packets
Nodes in the same DAS could fulfill batch backup of certain entries (e.g. ARP entries) with each other

9. Thank you Q&A