Stacking (IRF) two 1950 12XGT 4SPF+ switches (and a non IRF connection to a different 1950)

[Joker535]

I have two 1950 12XGT 4SPF+ switches. I bought those specific switches because I have 5 servers that I want to network using pairs of 10G connections (via LACP to get 20G). I understand that each 12XGT only allows 4 pairs of ports to support LACP at once (which is why I bought 2 of the switches giving me 8 pairs to work with). 

I started off with one 12XGT connected to 4 server each of which was using a pair of ports in LACP plus a 1950 24G 2SPF+ 2XGT that was just plugged into 1 regular port with a jumper (for iLo and other non 10G items). Everything worked great with this setup.

I added a 5th server and discovered the limit of 8 ports being used for LACP. I then purchased a second 12XGT and a pair of the stacking cables (3M SPF+ cables recommend by HPE support). I attempted to stack the two 12XGT switches using the IRF instructions from HPE (and instructions found elsewhere). I have not been able to get it to work and I can't figure out why. 

Can someone here link me to instructions that are correct for this process?

Is my 1950 24G 2SPF+ 2XGT being plugged in using a regular (non SPF) cable and not being configured for stacking causing a problem?

Thanks

[torinvoin]

in the last post there https://community.hpe.com/t5/web-and-unmanaged/stacking-irf-three-1950-switches/m-p/7078342/highlight/true#M4510 is a working instruction, just two weeks ago I made a stack on it.

[parnassus]

It's not clear: is your idea to have a server's LACP bond connected to both 12XGT IRF stacked switches concurrently? if so it's OK.

Say you will have:

Server NIC port 1 <--> interface 1 IRF-1
Server NIC port 2 <--> interface 1 IRF-2

Note: NIC ports 1 and 2 form Bond LACP exactly as interfaces 1/0/1 and 2/0/1 of IRF form LAG LACP.

A 3rd standalone switch connected to the IRF isn't an issue (I suppose you will connect it as you will do with the Server: LAG LACP to both IRF Members for redundancy).

[Joker535]

I was not intending for each server to connect to both switches. I was going to have both NICs on 2 servers connected to one switch and both NICs on 3 servers connected to the other switch. Does the stacking/IRF require me to conect each server to both switches instead of the way I have it? 

Would there be any advantage to running 3 of these switches stacked via IRF instead of 2 plus the 3rd switch that is different? I bought a spare 12XGT just in case we had a failure as we have already had 2 of them fail (replaced by HPE under warranty). I could add that to the stack and eliminate the 1950 24G 2SFP+ 2XGT.

Thanks

[parnassus]

---

@Joker535 wrote: Does the stacking/IRF require me to conect each server to both switches instead of the way I have it?

IRF lets you to do that...it's not that it requires you to do that...it permits you to do that (there is a big difference!).

Two Server NIC ports can be bonded in a LACP (IEEE 802.3ad) aggregate...that's quite normal...but the links MUST always physically - and concurrently - terminate to one physical switch (this is called "coterminus into a Switch")...or they must termiate into one LOGICAL switch...IRF makes your two 1950 Switches to act as a single logical entity...so having an IRF will let you to terminate one link to one IRF Member and the other link on the other IRF Member. You can't do that if you don't have a Virtual Switch(ing) solution like IRF. That's the "magic".

---

@Joker535 wrote: Would there be any advantage to running 3 of these switches stacked via IRF instead of 2 plus the 3rd switch that is different?

First of all you can't mix in a IRF deploymen two 1950 12XGT with any other model of the same HPE OfficeConnect 1950 Switch Series. So a 1950 IRF should be (a) made of all non XGT models or (b) made only of XGT models. You can't mix. If you have a 3rd 1950 Switch which is not a 12XGT you're forced to simply uplink it to the IRF, in other words you can't add it to the IRF stack already formed by two 1950 12XGT.

Given that...back to your question: why to use IRF and LAG LACP from peers? because of REDUDANCY this solution provides. If an IRF Member is lost (Fault) the other one continue to switch/route...so if you have peers connected to IRF via LACP...they will not suffer of any issue. The same is valid for any other switch connected to IRF through a LACP uplink.

[Joker535]

Thank you for that clarification. The redundancy issue may be a factor in this as we have already had two of the 12XGTs fail (one right away and the second after a very short time). The second one HPE replaced is now acting suspicious after only a few months of service,. I like having the the 24 port linked up as I can move all the 10G connections to it whent he 12XGTs fail (running at 1G without LACP but better than nothing). 

[parnassus]

What is the current software running version on the HPE OfficeConnect 1950 12XGT 4SFP+ Switch (JH295A)? is it the R5106P03 (03/2017)? unfortunately the JH295A software was developed less than the software used on all remaining models (see here for a comparison).

What do you exactly mean with "acting suspicious"? Just curious.

[Joker535]

I just looked in one of the switches and it is running R5103P03. I see R5106P03 is available. Should I upgrade or not?

The switch acting funny is taking a very long time to boot when plugged in. 

[Joker535]

Maybe one of you can clariify something for me. 

I want to stack two of the 12XGTs together. Each has 4 stacking ports. 

If I was to add 2 of the SFP ports (port 13 and port 14) to one IRF port configuration and the other 2 SFP ports to the second IRF port configuration on each switch, would that give me a 20G connection between switches (instead of the 10G)?