To LAG or not LAG?

Question

I have a new ex4200 switch stack, my 3 esxi hyper-visors connect to the SAN via vlan 200 which is EST. I have loaded the Equallogic multipathing plugins. When I have the 4 cables per server dedicated plugged in to the switch the hyper-visor see's 4 paths like it should. When I lag 2 ports and 2 ports together I only get two paths. This is expected behavior however what is better ? Use 2 2g links dictated by the switch or use 4 paths and let the plugin do it's job ?

score 5 · Accepted Answer · answered Jun 15 '11 at 15:41

5

With your SAN, let MPIO do its job and do not use LAGs with the server or SAN connections. You can (and should) use LAGs or LACP trunks between the switches though.

Here is the EqualLogic configuration guide with more information on this topic [Section 4.5]: http://en.community.dell.com/dell-groups/dtcmedia/m/mediagallery/19852516/download.aspx

answered Jun 15 '11 at 15:41

Doug Luxem

9,612
7
50
80

Second that, really! – 3molo Jun 15 '11 at 15:43
Multipathing does its job fine. – SpacemanSpiff Jun 16 '11 at 05:56
Also, be sure to add the no-split-detection statement to the virtual chassis configuration otherwise you're in for some fun in a failure situation. – SpacemanSpiff Jun 16 '11 at 05:57

score 1 · Answer 2 · answered Jun 15 '11 at 15:36

1

I would always use as many LACP trunk members as practical, not only for resilience but for potential performance benefits too, so I'd let the plugin do its job.

answered Jun 15 '11 at 15:36

Chopper3

101,299
9
108
239

What do you consider a better method; LAG or multiple individual links in conjunction with STP? I can see using LAG for increased throughput when the need has been qualified, but for providing redundant links between switches I usually configure multiple links and rely on STP to prevent loops. – joeqwerty Jun 16 '11 at 00:45
@joeqwerty- Unless I'm misunderstanding you, STP is going to put those redundant ports in blocking mode until there is a link failure. So in the case of a SAN using 1gb ports, your inter-switch connections would be a bottleneck with STP. However, in a more complex situation with many switches and trunks, STP would be useful. – Doug Luxem Jun 16 '11 at 13:05

score 1 · Answer 3 · answered Jun 16 '11 at 00:38

I am assuming that you are connecting your storage via iSCSI. If so, definitely let the MPIO do its thing, instead of aggregation. However, if you decide to use NFS, you have to keep in mind that with one datastore there will only be one TCP session and loss of that session means a loss of your datastore, so you are going to need something to give you added layer of redundancy. Datastore diversity is important with NFS, as this way you can distribute and reduce impact of failure. In cases where NFS is used LAG id a good option. Critically, far too often people forget that their single switch is a SPOF, and they do nothing about it. I am personally a firm believer in cross-stack aggregation, which eliminates the switch as a SPOF.

To LAG or not LAG?

3 Answers3