tame rover
Currently we have 10 nodes, A700,A700s,A800, and 2 pairs of C800. Because A700s is no longer supported on 9.16, we are planning on replacing it either by A800,A900,or A1k. Using A800/A900 may not have questions. But, can we use Cisco n3k for A1k disk shelves connections in addition to cluster network connections?
If yes, how many n3k ports needed for disk shelves connections ? Can current 9.14 work with A1k?
velvet maple
You’ll need to be more specific
The 3232 can be used for storage. However: only for storage. The 3232 does not have a shared rcf config. It must be used as storage only or as cluster switching only.
I am petty sure all those platforms are compatible at 40/100g but require the correct rcf meaning you can use a pair of 3132 or 3232 or 9336 switches for cluster networking the newer platforms (c800/a800/a1k/a90 -> i don’t think you meant a900 since I think it is end of availability)
With all that hardware, you should purchase (from Netapp) a pair of 9336 switches. If you go that route then you can use one pair of switches for clustering and the same pair can support I think up to 8 shelves on two different he a-pairs.
Just for reference, when using switches for storage, there is no isl for the storage part. The groupings are isolated and the storage vlans are not allowed over the switch isl.
tame rover
Very helpful.
If we will have to upgrade switch for a1k, that’s gonna be a lot of investment at one time. Will a90 require the upgrade as well? What differences in capacity ,cpu/memory/storage between a1k, a800,a90 ?
velvet maple
That I would refer you to hwu!
https://hwu.netapp.com/DetailViews/OverviewPage.aspx?platformTypeId=5265148
If you are already in with the A1k, the incremental cost for the switches should only be within a single digit percentage of the overall cost
Plus, your Netapp team really REALLY should have brought it up
The new platforms (a90/c80/a1k/fas50 as examples) all use a single wire for ha+cluster. It is also possible to connect to switches badly such that ha doesn’t work. The ha vlan being used stays local to the switch and doesn’t go over the isl. if the ha pairs cannot talk to each other, ha connectivity fails.
With that, the 3k and the 9k both work but the 9336 is just better over all since you can do cluster AND storage with the same switch pair for up to two ha pairs (for storage) and supports up to I think 10 high speed (100g) nodes
tame rover
The new platforms (a90/c80/a1k/fas50 as examples) all use a single wire for ha+cluster.
I am not familiar with how these work togetehr, how can I understand "single wire"?
Will cluster networking be used for HA as well?
Doesn't storage and networking use different cluster switch ports?
tiny verge
they use vlans
velvet maple
A "node" port uses this:
description 10/25/40/1000GbE Intra-Cluster/HA Nodes
switchport mode trunk
switchport trunk allowed vlan 1,17-18
spanning-tree port type edge trunk
spanning-tree bpduguard enable
mtu 9216
state enabled
exit
the ISL uses this
description LOCAL_ISL Switch
service-policy input ROCE_QOS
priority-flow-control mode on
switchport mode trunk
switchport trunk allowed vlan 1
mtu 9216
no shutdown
Notice the Port-channel does NOT pass anything but VLAN1.
Node HA-A (example e0a) will use VLAN-17
Node HA-B (example e0B) will use VLAN-18
If you cross the HA (put node-01:e0a and node-02:e0b on the same switch), then HA will not be able to communicate
VLAN 1 is (the 169.254.0.0/16 addesses) is the Cluster Network and passes allowing Full-mesh IP connections between the nodes.