#Intercluster lif port recommendations.

1 messages · Page 1 of 1 (latest)

thorn fossil
#

As the title says,

I've got a need to jigger around my intercluster lifs some.

The ports they live on are 2 multimode_lacp port channels consisting of 10GbE pairs.
Via harvest, I've validated that one our of two of those ports is getting nailed to the wall during peak periods for replication, front end traffic, and fabricpool traffic.

The simplest way for me to alleviate this (I believe) is to just add more ports and move intercluster traffic to those new ifgrps/ports.

My question boils down to this, is there a better way on array to balance out traffic on these ifgrps for snapmirror and fabricpool? As it appears sessions are being stood up on primarily one and not going anywhere from there...

Our ifgrps are currently using IP as their distribution method, should I target sequential to up the utilization? Or for snapmirror // fabricpool will this not make much of a difference?

solemn dirge
#

ONTAP should balance the traffic when you create say 2 IC LIFs per node. Traffic should flow equally through both. As long as you have only one LIF on the ifgrp, it will never be faster than a single member because it's a single stream

#

if you use port-based ifgrp it should also be possible to increase the bandwidth, but that's still not a guarantee that all links are loaded equally

#

TR-4015 has some discussion about the multipathing details starting at page 13

thorn fossil
#

Port based, that's a good recommendation.

Does that take into account fabricpool traversing the intercluster lifs? Or just snapmirror?

#

I'd presume just snapmirror with the TR it's in

solemn dirge
#

it depends on whether you use separate ipspaces for snapmirror and fabricpool. if it's the same ipspace, then it potentially also affects fabricpool (i.e. you might see S3 traffic from the "new" LIFs to your S3 as well)

#

the problem is that you cannot "switch" the ifgroup to use a different balancing function, you have to delete and re-create it. this is disruptive of course (unless you have free ports and can migrate the LIFs somewhere else during the transition)

thorn fossil
thorn fossil
ornate dust
#

IP distribution is not very good. I always use PORT distribution. Unfortunately the only way to change that is to delete and recreate the ifgrp

thorn fossil
ornate dust
#

In a cluster, you can do one node at a time. You can likely do it non disruptive but likely only if you are not tiering (fabric pool).
Re-home all lifs on the ifgrp to another node. If you have snapmirror, you probably know you cannot move intercluster lifs to another node.You would need to delete the intercluster lif. Delete the vlans Now you can delete the ifgrp, and recreate. Then add vlans back and recreate the intercluster lif. Recheck cluster peering.

It’s a process. I’ve done it a bunch of times for customers

solemn dirge
#

if you have easy access to the switches you can potentially do it nondisruptively: remove one port from the ifgrp, remove it on the switch as well, then create a new ifgrp (in ONTAP and on the switch) with that one port, move all LIFs over to the new ifgrp, then delete the old one and add the second port to the new ifgrp. Much more involved but keeps your intercluster LIFs online

thorn fossil