#Downgrade cluster connection from 40GB/s to 10GB/s in the backup site

1 messages · Page 1 of 1 (latest)

halcyon crag
#

Hello Folks,

In our current environment in the backup site, we have 2 x FAS 9000 and a AFF-A700 controllers. There's a need to downgrade infrastructure from 40GB/s to 10GB/s connection. Below is what we have in terms of hardware specs:

Current active connection: 10/40 gigabit ethernet controller XL710(0x1584) QSFP+ providing high performance.
Looking to use: 1G/10G Ethernet Controller CNA EP 8324N (Qlogic EP8324 Quad-Port 16GB FC/10GbE PCIe CNA). Already have card available in slot 1

what's the best or easy approach to go about
getting it changed? We want to decommission 40GB cluster switches completely and want to use 10GB switch that we’ll install and cable it up

thorn zinc
#

this is not supported. You cannot use the X91143 (UTA) card for the cluster network. It has to be the 40 gig card (X91440) in slot 4, although you can use 40g-to-4x10g breakout cables with that (although I would not recommend it).

Also, according to hwu.netapp.com, the only supported backend switch for the 10g configuration is the BES-53248 (for the A700), so you couldn't replace that one anyways....

Since the cluster switches are dedicated to the NetApp controllers anyway, I wonder about the reasoning behind the downgrade...

daring halo
#

This is over on reddit also.
If you have 2 x X91440 in your controllers, leave them and use e4a and e8a for the cluster network connections for best performance. If you choose to want more 10G ports, you can flip ports e4e and e8e to breakout (and get e4e,e4f,e4g,e4h and same with slot 8). These are ethernet only. Pretty sure this is the ONLY platform where you can breakout the card this way (other platforms restrict breakout to port a only and then disables port e). You would need to power cycle the node to get th breakout (not just a reboot on this platform)

halcyon crag
#

Ok guys. Thanks for the info and update. Much Appreciated!

silent hinge
#

Why downgrade to 10 Gb/s for cluster network?

daring halo
#

There are cases it makes sense. When you have one beefy node (fas9000) and a bunch of smaller nodes (fas2700/a300) where the majority of nodes only support 10g. It makes inter node communications more efficient

#

Generally I don’t worry about it