SG350 Stack topologies question

[ArnaudG]

Hello, I am wondering what is a valid/optimal scenario in the current topology I am working on. 

I have 3 stacks of switches

Stack SA :  2x SG350XG-48T

Stack SB : 2x SG350XG-24F (core stack) 

Stack SC : 3x SG350X-24

LAG questions 

- which LAG type is preferred ? Etherchannel ? other ?

- are there some specific parameters to apply to optimize bandwidth and latency ?

- in the topology B, should I create one big LAG containing 6 links or one LAG per switch with 2 links ?

Topology 

- Is topology A valid ?

- I understand topology B will upgrade redundancy for sure. but what I am not sure about is how the packets will actually travel the links. 

e.g. H1 wants to send Hello to H2

On topology A, I understand the shortest path is SA1 > SB1 > SC1 > SC2. but I am not even sure my packet will really take this optimized path... 

On topology B, I can theoretically have one less hop because of the SB1 to SC2 link. But will Etherchannel and stacking actually optimize this way ?

 

[Reza Sharifi]

Hi,

Not familiar with SG series, but option A should work fine since all switches are stacked in each layer.

In the Cisco world, LAG and Etherchannel are the same.

So, for option-A, put the 10gig links from SC1, SC3, and from SB1 and SB2 all in one Portchanel.

Create a second Portchannel and put the links from SB1, SB2, and SA1 and SA2 in it.

HTH

[Joseph W. Doherty]

"are there some specific parameters to apply to optimize bandwidth and latency ?"

Usually not.

Stacking technologies have their only proprietary (i.e. often not detailed by vendor) ways of using their redundant paths.

So, for example, in your B topology, SB1 "sending" to SC1, even though there's a direct link, might first go "sideways" via SB2, and while doing so, we might be unable to predict which stacking cable between the SB switches will be used.  Further, when data arrives on SC1 the stack might send the frame via SC3 even though the host is connected to SC2.

Regarding Etherchannel (i.e. LAG) links between switches we also don't much optimize bandwidth and latency.  Generally we work to spread the load, evenly, across all the links.

With Cisco Etherchannel, how flows are distributed depends on how the attributes of the frame or packet "hash" out across the member links.  Worst case, only one link is ever used.  Best case, might also only use just one line, for example, like your case between just a pair of hosts.

I don't know about SG switches, but on many Cisco switches Etherchannel works best with powers of 2 number of links.  Further, for performance, the "next" better interface, in bandwidth, is often (sometimes) much better than an Etherchannel with maximum number of member links.

[ArnaudG]

Hello ! 

Thanks for your replies.

My understanding is that option B will only guarantee an higher level of redundancy.

- Bandwidth will not be optimized by the LAGs because capped in any case by the 10G native stacking.

- Moreover as the packet route between two hosts is not following necessarily the shortest route, there will be no gain regarding latency.

I guess that I'll stick to topology A which is redundant enough and clearer in its implementation. 

Unless someone would expose other options

[Reza Sharifi]

 Bandwidth will not be optimized by the LAGs because capped in any case by the 10G native stacking

That is correct. Remember a flow can only use one of the physical interfaces in a given Portchannle no matter how many links you have in there. In the case of muli 10Gig, the max is 10Gig.

HTH