Interconnect 2 IRF Stack (LCAP protocol)

[yfourar]

Hello,

I have two models of HPE 5130 EI and HI switches, I would like to mount two stacks (Stack 1 contains 3 HPE 5130 HI switches and the 2nd stack contains 6 HPE 5130 EI switches) and interconnect them via DAC cables (3 links), but it will cause a loop problem, so I thought to activate the LCAP protocol between both stacks, so my question is is this the right solution? and do you have other proposals.

Thank you in advance.

[Ivan_B]

Hello @yfourar !

If all 3 links will be in the same link-aggregation group (BAGG, Bridge-Aggregation), then there will be no loop, as all 3 links will be treated as 1 single logical wire. LACP (dynamic link-aggregation) is a good choice and it is much safer than static link-aggregation.

[yfourar]

Hello again, 

Basically, I would like to build this architecture 

so will there be any problems?
And yes all the 3 links will be in the same link-aggregation group.

Thank you.

[Ivan_B]

I see nothing that may prevent you from using this topology. I have only one concern, maybe even advice. After implementing this topology you need to monitor utilization of the following IRF ports:

Since 3 of 6 members of the lower stack won't have any links to the upper stack, in case they will have any traffic for the upper stack, they will need to push it through the IRF port to the closest IRF member which has port in the BAGG. If you see overutilization of those IRF ports, consider two possible solutions:

1. Increase number of IRF links to increase overall IRF logical ports bandwidth.
2. Connect lower 3 members to the upper stack as well, so each member in lower stack will have at least one link to the upper stack:

But again, it's just a general thought, all decisions must be taken basing on facts about actual data flows, link utilization etc...

[yfourar]

So according to you, the use of the architecture that you sent me is more reliable and relevant, Ok I will use it. Thank you.

Another question, for the LCAP link, I use the fixed SFP ports?

[Ivan_B]

You can safely start with 3 links as planned and then check the utilization of IRF ports and add more links either to IRF or to the link-aggregation. I hardly can advise one or another scenario without deep analysis of particular network and traffic flows. But having all IRF members participating in such link-aggregation seems more efficient than initial plan. However, it doesn't eliminate eventual need of expanding IRF ports capacity, again, it all depends on actual traffic flows.

Being OSI Layer 2 protocol LACP is not physical media type dependent and definitely works on SFP+ ports no matter what transceivers you use there - copper, optical, it should work on all kind links that work according Ethernet standard.

[yfourar]

it seems to me that it's the best solution.

If I add IRF links, I will remove the LCAP links, or the opposite (in my oppinion I'm using ring topologie it is more reliable than the Daisy connection, so even if there was a failure of a link it will not affect the IRF service) so do you agree with me ?

Another question please ! what did you mean about checking the utilization of IRF ports and Can I add IRF links ? Yes, but the problem is that between (the second & the third) / (the fouth & the fifth) I can't  because all of the ports are used. (see the architecture) 

thank you.

[Ivan_B]

By checking the IRF links utilization I mean you need to check how busy they are. Example:

If IRF port is Ten1/0/50, then you can 'display interface Ten1/0/50" and check following lines:

Peak input rate: 148 bytes/sec, at 2013-01-01 00:06:04
Peak output rate: 40 bytes/sec, at 2013-01-01 00:11:55
Last 300 second input: 1 packets/sec 145 bytes/sec 0%
Last 300 second output: 0 packets/sec 40 bytes/sec 0%

In my example you barely see any traffic, but if 'Last 300 seconds" counters on your switches start to go over 60%, you can start planning adding new physical link to logical irf-port. Like in your last example where members 1 and 2 have two physical links between ports 49 and 50, but both links are member of one logical irf-port.

Regarding your last topology - you can start with just single physical link between each IRF member and then if you see high utilization on any IRF port then you can add physical link to that logical IRF port. Example - in the beginning you have 1 physical link between member 1 (master) and 2 on port 49:

irf-port1/2
port group interface Ten-GigabitEthernet1/0/49
#
irf-port2/1
port group interface Ten-GigabitEthernet2/0/49

After some time you see that the utilization of that link (last 300 seconds input/output) keeps steady above 60-70%. Only then you add second physical link to the irf-port1/2 and 2/1:

irf-port1/2
port group interface Ten-GigabitEthernet1/0/49
port group interface Ten-GigabitEthernet1/0/50
#
irf-port2/1
port group interface Ten-GigabitEthernet2/0/49
port group interface Ten-GigabitEthernet2/0/50

But if all ports are busy on a switch, like on 4th ... then we lack this possibility to expand IRF's bandwidth.

That is why start with single link and add links only to busy segments. For example maybe link between slot 1 and 2 won't be busy, but 2 and 3 will be much more utilized. Then you won't need that second link between 1 and 2 and better use free ports to expand the busy segment between 2 and 3. 

[yfourar]

Can you tell me what is the best architecture I should use. the 6 switches stack i will put the users and the printers and, the other stack(3 switchse) i will put the servers
Ps. I will create the vlans too (users, serverus, printers)