Backplane Traversal

[AllanHedegaard]

AFF200 Active/passive, with one SVM on the active node, that also owns the data aggr.

Say I put a SVM lif on the passive node. 10Gbe SFP+. How will that affect performance? How high is the throughput of IC? I am running switchless. The access protocol is nfs.

Thanks

[GidonMarcus]

Hi

Unlike 7 mode (for FCP only as well) the data for the incorrect path is not going on the IC/PB. it goes via the cluster network, by default on the AFF200 it's ports e0A and e0B which are 10GB each  (can also use other ports - on the same speed).

it should add very small latency (around 1ms).

as NFS v3 is stateless. i would place the LIF on the same location as the volumes. and let the LIF failover to handle redundancy. in NFS v4 or if pNFS involved - need to read some docs, but in general from my experience the LIFS fails-over faster than the aggregates... (assuming ARP cache on the switches not creating any problems).

Gidi

[aborzenkov]

---

@GidonMarcus wrote:

it should add very small latency (around 1ms).

---

This is extremely HUGE latency for an all flash array.

[AllanHedegaard]

Know the saying - Never assume it's a one way street?

I just tried running a 4k 100%  random  100 % read on a IOmeter load, while failing over the LIF to the other controller.

Either the 10Gbe network is the limiting factor or the indirect connection is not being affected by the failover.

24x 960GB SSD