LACP LAG Load sharing methods

[treese]

There are many load sharing methods with LAG for Extreme and I would like to see documentation or have someone explain each one. Reason being is we have LACP LAGs, MLAGs, static LAGs in our environments. Some are Extreme to Extreme and others are FW to Extreme, Linux to Extreme, ESX host to Extreme. Just looking for the best load distribution and have always done address-based L3_L4 LACP. I believe our single port LAGs to MLAG set is this way too.

Thank you, very much.

[treese]

Heck I forget, its been so long. If I have a LAG with L3_L4 and LACP set but the server is only set in a LAG with LACP, would this have better load distribution then L2 LACP LAG?

[Patrick_Voss]

Ted,

I am not sure what you are asking here. Unless you are referring to the Static portion. Static basically means it is a lag without any link to link protocol (like LACP). If my assumption is incorrect please paste the outputs you are referring to.

[treese]

I've noticed if I config a edge switch with a shared ports 47 & 48 then hash configured for address-based L3_L4 and then the uplinks are the MLAG ports on each core shows to be a static LAG. No sharing on those single port uplinks.

If I took the MLAG ports and make them address-based L3_L4 then it changes. Its a single connection to each core but essentially 2Gig on the edge switch side. I'm not sure what I'm gaining by configuring it like scenario 1 or the later. Both have active active connections one to each core.

[treese]

Thanks for the explanations, and this helps. Thank you.

[Paul_Thornton]

This actually breaks down into three questions/clarifications - all of which are independent of each other.

Firstly: What type of LAG do I have? Is it single-switch to single switch (or single stack to single stack), so a normal common-or-garden LAG (port channel in Cisco speak). If it is from two switches to a third switch, then it is MLAG.

Then: What hashing algorithm does the sending switch use to decide which port to transmit the frame from? If it is L2, the only the MAC addresses are used; if L3, then the IP addresses are used; if L3_L4 then both IP addresses and TCP/UDP ports are used. It depends on your traffic type to a degree, but you generally won't go far wrong with L3_L4 in any normal environment using routing and a reasonable number of clients. It isn't a problem having different hash algorithms at each end of a link - but you don't normally do it as that wouldn't lead to the best load balancing between them.

Finally, there's the question of LACP. Do I want an active protocol to control whether or not a physical link between two devices can be used to carry traffic as part of the LAG. I am a firm believer in always using LACP as it saves you from some issues with blackholed traffic - however, others think that this introduces too much complexity and that static LAGs are the way to go.

Remember also that it doesn't really matter what you have on the other end - another Extreme switch, a Cisco switch, a Linux host or a VMWare cluster - so long as both ends are aware that there is a LAG, and if LACP is enabled on one end, it must be enabled on the other or you won't get anything going across the links.

Paul.