Are there any use cases for interface groups?

[TAO1]

Hi,

Given the advent of 10GbE+ capability and LIFS which can fail-over, assuming you don't have a workload which busts the limit of a single Ethernet port, are there any use cases for, or advantages to using interface groups?

Thanks

[aladd]

It would negate the need for LIF failover, additionally some throughputs are enough to overwhelm even a 10GBE connection.

[TAO1]

Thanks. So other than failover and bandwidth, no other advantages/benefits/features? Would any particular protocol reap the benefits of using ifgrp?

I guess you could argue that a port failure in an ifgrp causes a failover event.

[parisi]

Stateful protocols like NFSv4 and SMB would benefit from ifgroups, as it would minimize the disruption in the case of a port failure.

[TAO1]

Are there any documented resources which outlines why this would be the case please?

[paul_stejskal]

I can look, but I think it is because CIFS is stateful and doesn't like to be disconnected by default. I don't think we have anything saying specifically ifgrps are better for stateful protocols.

[parisi]

For recent versions of SMB, it's not as useful. SMB2 and 3 have features such as durable handles that help mitigate impact from network issues.

SMB1 benefits greatly from ifgrp failover, but SMB1 is being deprecated.

NFSv4.x would benefit similarly to SMB1, as its features for network disruptions aren't as robust as current SMB iterations.

But as @paul_stejskal said, no official docs on this.

[TAO1]

Thanks for the reply.

I understood that traffic between a source and destination will use the same physical port in an ifgrp for the same session, i.e. the traffic between a single source and destination doesn't load-balance across all active ports in an ifgrp. Balanced load across an ifgrp occurs where you have multiple sources of traffic  to the ifgrp (my understanding 😊). If the physical port which is carrying the session fails then there would in effect be a failover event to a surviving physical port in the ifgrp.

If my understanding is correct (?) then would you say that, for NFSv4 for example, ifgrp failover is less impactive than a lif failover (with no ifgrp), however there is still impact?

[TAO1]

Re my above port, the comment is assuming dynamic multimode ifgrp. Any thoughts please? 

[Tom_Holz]

I started yesterday a training about Ontap Cluster Fundamentals, talking exactly about the ifgroups. 

They make sense in order to organize the connections in a more clean way, and give the option to share on IP to many port. A nice external source states: 

>>NetApp Interface Groups are similar to NIC teaming on a standard server. They can be used to achieve redundancy (and possibly load balancing) for our network ports, links, and switches. NIC teaming bundles multiple physical links into a single logical link with one shared IP address.<<

from: NetApp Interface Groups Tutorial - FlackBox

[TAO1]

Thanks for the reply Tom.

I guess beauty is in the eye of the beholder, to my mind they don't organize the connections better - they add another layer of complexity. Just my personal opinion.

Just to restate my query, if your workload fits within a single physical port bandwidth capability and lifs provide failover capability, what's the advantage of ifgrps?

Parisi mentioned NFSv4 would benefit in failover, however given that both LIFs and ifgrps (in some circumstances) would have to failover sessions to different physical ports, I don't get the benefit. 

Thanks again

[parisi]

Session states are kept per node, so a port failover shouldn't be impactful on the same node.

And for ifgrp, the MAC that gets sent is a generated MAC address based on the ifgrp rather than the MAC of the actual physical ports. 

https://kb.netapp.com/Advice_and_Troubleshooting/Data_Storage_Software/ONTAP_OS/How_does_a_VIF_MAC_address_get_assigned_in_ONTAP_7%3F

But if you don't need the extra stuff, then just use single interfaces. 

[TAO1]

"Session states are kept per node, so a port failover shouldn't be impactful on the same node." - so that goes for igrp and lif port failover?

"And for ifgrp, the MAC that gets sent is a generated MAC address based on the ifgrp rather than the MAC of the actual physical ports. " - sorry, I'm not entirely sure what you are driving at here. Are you saying that a LIF failover has a disadvantage over ifgrp because a LIF needs to 're-bind' itself to the MAC address of the port that it is failing over to?

Thanks