Within the first weblog publish of the MLAG Know-how Deep Dive sequence, we explored the parts of an MLAG system and the basic management aircraft necessities.
This publish focuses on a serious constructing block of the layer-2 information aircraft performance: MAC studying. We’ll maintain utilizing the identical community topology with two switches and 5 hosts, and assume our system tries its finest to implement hot-potato switching (sending the frames towards the vacation spot MAC handle on the shortest doable path).
Easy MLAG topology
Think about A sending frames to X over the A-S2 hyperlink. S2 forwards (or floods) these frames over the S2-S1 hyperlink. Utilizing the normal MAC handle studying, S1 would deduce that it ought to use the S1-S2 hyperlink for the return site visitors towards A, violating the hot-potato switching expectation.
Conclusion: Switches in an MLAG cluster shouldn’t rely completely on dynamic MAC studying. They want a control-plane protocol to alternate MAC addresses reachable over the LAG member hyperlinks. It’s price noting that ICCP (RFC 7275) doesn’t present this performance; to construct a standards-based MLAG answer, you must mix ICCP with one other control-plane protocol like EVPN.
Can we mix dynamic MAC studying on the S1-S2 hyperlink with a control-plane protocol? It’s a tricky problem. Think about A utilizing each LAG members to ship site visitors to X; S1 would see supply MAC handle A coming from the directly-connected hyperlink and the S1-S2 hyperlink. Conventional MAC studying conduct would shortly set off duplicate MAC handle alerts; clearly not one thing we wish to see.
You’ll be able to clear up this conundrum in a number of methods, relying on what the underlying {hardware} helps:
- Implement an handle supply precedence scheme – MAC addresses marketed by the management aircraft can’t be modified by dynamic MAC studying.
- MLAG implementations utilizing a peer hyperlink normally use proprietary encapsulation on the peer hyperlink. The change might use the “this body is coming from a LAG member” a part of that encapsulation to affect dynamic MAC studying.
- It is perhaps best to surrender, disable dynamic MAC studying on the peer hyperlink, and use a control-plane MAC handle commercial protocol between MLAG cluster members.
I’m constructive we are able to discover implementations of all three concepts within the wild; if you know the way a selected MLAG implementation works, please go away a remark.
