HPE 5130 EI JG934A - LAGs number

[Skyscraper]

Hello,

I'm considering buying 4x  HPE 5130 EI (JG934A) and putting them in IRF stack.  I haven't found exact information about numer of LAGs(LACP) supported by the switch to be able to connect to servers (bonding/teaming) and other external siwtches (LACP)? This information is very important for me and I'll be gratefull for Your help. Thank You.

Janusz.

[parnassus]

Generally the number of aggregable interfaces on a single LAG is limited - on a single device - to 8 and this value should be valid for the HPE FlexNetwork 5130 EI Switch series too.

For the same series, the maximum number of inter-device LAGs should be limited to 128 (so when LAGs are spread across the entire IRF stack with member interfaces belonging to different IRF members); there should be also a "per device" LAG limit of 14 (so at most 14 LAGs on a single device).

[Skyscraper]

Thank You for Your answer. To be sure please verify if I understand You correctly.

The situation is I have 2 switches 5130 as members of single IRF stack., and I have 40 servers (each with 2 teamed -lacp- NICs in). And each server is connected to IRF as follows:

NIC1<->SW1 (1st IRF Member)

NIC2<->SW2 (2nd IRF Member)

Will I be able to create 40 LAGs/LACP on IRF for each server? I don't catch the "14 LAGs per device" idea :]

Thank You in advance.

[parnassus · Accepted solution]

You should be able to create 40 LAGs, each one made of a two member interfaces, spanning the two members IRF following this possible schema (numbered that way to be easy to troubleshoot):

Server n NIC Port 1 <-> Interface 1/0/n (IRF Member 1) member of LAG n
Server n NIC Port 2 <-> Interface 2/0/n (IRF Member 2) member of LAG n

where n=1, 2, 3, ... up to 40.

I think the limit of 14 LAGs was cited to make evident that if you have a 8 or 9 members IRF...considering the 8 members per LAG limits...then you can only have at most 14 LAGs made of 8 member interfaces (one per IRF member) on the whole IRF...reaching something like 112 interfaces aggregated into 14 LAGs.

It looks like (Release Notes) up to 128 LAGs are permitted on (most extended) IRF deployment.

[Skyscraper]

Thank You very much for Your clarification, I appreciate it a lot :)

It seems that in 2 member IRF (2x 48ports) I can fill up all devices with 2 port LAGS easily.

Would it be true, in case if I have 4 member IRF (4x 48ports), to fill up all 4 devices with 2 port LAGs using following configuration?:

Server Y NIC Port 1 <-> Interface 1/0/n (IRF Member 1) member of LAG Y
Server Y NIC Port 2 <-> Interface 2/0/n (IRF Member 2) member of LAG Y
where n=1, 2, 3, ... up to 48.
where Y=1, 2, 3, ... up to 48.

Server X NIC Port 1 <-> Interface 3/0/n (IRF Member 3) member of LAG X
Server X NIC Port 2 <-> Interface 4/0/n (IRF Member 4) member of LAG X
where n=1, 2, 3, ... up to 48.
where X=49, 50 ... up to 96.

[parnassus]

Yes, it should be possible (LAG members can be part of any IRF member).

For IRF Member 1 and 2:

Server X NIC Port 1 <-> Interface 1/0/n (IRF Member 1) member of LAG X
Server X NIC Port 2 <-> Interface 2/0/n (IRF Member 2) member of LAG X
where n=1, 2, 3, ... up to 48.
where X=1, 2, 3, ... up to 48.

For IRF Member 3 and 4:

Server X NIC Port 1 <-> Interface 3/0/n (IRF Member 3) member of LAG X
Server X NIC Port 2 <-> Interface 4/0/n (IRF Member 4) member of LAG X
where n=1, 2, 3, ... up to 48 (as above since you have IRF Members with at most 48 interfaces)
where X=49, 50 ... up to 96 (correct).

De facto you're splitting half of your Servers' LAGs on the first IRF pair and the remaining half on the second IRF pair. Clearly it's totally possible to adopt other LAGs links "distribution" scheme against your entire IRF stack (you should consider where your Servers' traffic flows go to better try to balance the resulting inter-IRF traffic).

[parnassus]

With more than 2 IRF Members you should carefully consider to deploy a IRF Ring topology where each IRF port is made of at least two physical interfaces (for resiliency and load balancing)...and, probably, then you would alternate Servers' LAGs terminations on non consecutive IRF Members (It depends...).

[Skyscraper]

Yes, about Ring IRF topology - this is exactly what I Was planning to implement. The idea is to make 4 member IRF with 2 10G SFP ports used on each device. Devices connected in Ring Topology using DAC cables. Then each server with 2 NICs as LAG/LACP  splitted and connected to separate IRF Members.

After using 2 ports for IRF stackiing on device (JG934A), can I use the other 2x  10G SFP ports as independednt uplinks? Or are they reserved for IRF ?

Thanks again for Your involvement :)

[parnassus]

Consider that JG934A has four SFP+ interfaces on the front panel, these four SFP+ interfaces are grouped in two groups:
(a) SFP+ ports 49 and 50 in one group
(b) SFP+ ports 51 and 52 in the other group
Physical member interfaces of one logical IRF port must belong to the same group (that's to say that you can't - as example - bind interfaces 49 and 51 to the same IRF logical port).

I can't find a explicit restriction about using remaining free SFP+ interfaces in the range 49-52 for other non IRF related uses (such as one aggregated links downlink or two single link downlinks) so - IMHO - if it is not denied, it's permitted.