#Cluster and Data Network

1 messages · Page 1 of 1 (latest)

celest hornet
#

Suppose we have 4 node cluster ( it May be FAS or it may be AFF) and now we want to connect this all 4 nodes through cluster interconnect switch so:

  1. Which switch can we use to connect these all 4 nodes? (LAN switch or SAN switch)

  2. If LAN switch then why LAN switch? and if SAN switch then why SAN switch?

Now we have connected 4 node though cluster inter-connect switch so can we use same switch for the Data network??

  • if yes, then suppose cluster-interconnect switch is SAN and customer wants to use ethernet protocol so do we also need to have the LAN switch or can we use same SAN switch? Because as per my understanding we can not use SAN switch for ethernet connectivity (Plz Correct me if I am wrong)

  • If yes, then suppose cluster-interconnect switch is LAN and customer wants to use FC protocol so do we also need to have the SAN switch? Because as per my understanding we can not use LAN switch for FC connectivity (Plz Correct me if I am wrong)

wraith garnet
#

No. Cluster network switches are supported for cluster expansion and storage attachment only. It is not supported to share with front-end networking.

grim thicket
#

Don't even think about using the same switches for both functions. One pair of switches for cluster back-end and one (or more) pair(s) for front-end data.

fading sorrel
#

and only a handful of switches are actually supported for the process. Current shipping switches include: BES-53248, Cisco Nexus 9336C-FCX2 and the NVIDIA S2100.

#

As @wraith garnet indicated , the switches are dedicated to cluster networking. Not supported to use them for front-end (user) networking at the same time.

grim thicket
#

There are good reasons why only a handful of switches are supported and even then, weird switch connectivity issues can silently corrupt your data. I've got a few BURTs out there for that (not all fixed yet) and I had those issues on SUPPORTED switches. If you want to keep your data intact, don't fool around with unsupported configs on the back end. Cluster switches should be cheaper than the cost of your data. Trust me when I say that silent data corruption is not any fun to find or fix. And by silent, I mean no error messages to the clients - you just get back different data than what you wrote.

sand patio
#

Plus we in Support will tell you to get into a supported configuration, and may not even be able to provide support or refuse to.

humble basin
grim thicket
humble basin
#

yeah, bad cables can definitely ruin your day. OTOH, we are running a handful of semi-productive lab clusters on an unsupported switch for years, and never had any issues like that..... it all depends, and the general rule always applies: if you want support, use a supported config

#

but fun fact: the cables and SFPs that NetApp sells and uses in the Cisco backend switches are not supported by Cisco, and if you have a NetApp support case that involves Cisco in the backend, good luck finding a few Cisco-supported SFPs/cables somewhere that you can swap in 😄

#

it's not easy telling your customer that "sorry, the SFPs that we plugged into your Cisco switches, we need to replace them with Cisco SFPs for your NetApp case to go on"

quasi mantle
#

@humble basin so why not deploy the product with the Cisco SFPs?

humble basin
quasi mantle
#

Ahh, now it makes sense(not really) but why does NetApp supply… forget it, enough ignorance shown already.

willow bay
#

...just a quick question about this... we have some CN1610 collecting dust. And we might want to expand a HA setup of FAS27xx nodes into a 4 node custer. but I think the CN1610 was dropped from ONTAP 9.12 onwards? But how "dangerous" would you rate it to use these switches in a setup? Bare in mind that this is mainly a snapmirror destination with limited support anyway...