NFS High availability

[#1]

Hello!

I'm looking to implement a FAS2020 device as shared storage for three Citrix Xenserver hosts. My past experience with storage has always been Fibre Channel, and I already have in place a small "infrastructure" for that - two separate fabrics each consisting of a HP 4/8 Base SAN Switch, some dual-HBA servers with one HBA connected to each switch and an HP MSA1500 dual-controller storage array with one controller connected to each switch.

My initial thought was to connect the FAS2020 to the same FC switches in the same manner, but having surfed around a bit I saw a lot of talk about why NFS might be a better solution. I'm almost sold on it, but I have no experience with NFS and Netapp, and I can't quite figure out how I would achieve similar high availability (no single point of failure) with NFS as I now have with FC.

So far I have thought the following:

First of course I must have a FAS with two heads (active/active). Then I should probably have two dedicated Ethernet switches. Each head of the FAS should be connected to both switches. Each server needing access to the storage should also be connected to both switches.

But I can't find any clear documentation about what to do next. How to configure the NetApp so that in case of one controller or one network switch failing the hosts would maintain access to storage?

I've read TR-3732 which talks about creating dynamic multimode VIF and using LACP, but this sounds like using a single network switch (=single point of failure). Is there any documentation describing the setup I'm thinking about, or am I way off course here?

[#2 · Accepted solution]

If you have two seperate switches (as opposed to a "huge" redundant single switch) you need to set up a single mode VIF because MultiMode requires you to have a trunk/etherchannel on the switch ports which you can't do across switches.

If you're worried about performance you can do two MultiMode VIFs, one to each switch, and set up a single-mode VIF ontop of that. You need 4 network ports for that (or 3 if you're ok with an asymmetric setup -- 2 primary and 1 backup path)

Then you set up the filers to take over each partner vif during takeover.

Then you only need to make sure that your Xen host has redundant network connectivity. On ESX you would configure the vSwitch where the VMkernel port is connected in such a way that it has at least 2 physical uplinks. I guess there's a similar functionality for Xen.

-Michael

[#3]

Since my environment is rather small, I'm more concerned about reliability than performance. Looks like I have a bit of reading to do concerning VIFs, but if a single mode VIF can have sub-interfaces (or whatever they are called) connected to different switches, then this seems to be the answer to my problems.

[#4]

Single vif mode is a failover mode, failover is completed managed by the NetApp controller (no particular switch configuration is needed) So your can connect the different links of a single vif to different switches without any kind of problem.

[#5]

Thanks a lot for your clarifications, people. Now I understand what I need to do.

[#6]

We have two switches, an 'A' switch that's the primary and a 'B' switch. The switches are trunked together (so anything using the B switch can see hosts on the A switch and vice versa). We have an rc file on our filer that looks something like this. The interfaces 'e0a e0b e0e e0f' are connected to the A switch and 'e0c e0d' are connected to B, as our standby. Cisco etherchannel is used on the switch side.

from /etc/rc:

vif create lacp vif-a -b ip e0a e0b e0e e0f
vif create lacp vif-b -b ip e0c e0d
vif create single vif-master vif-a vif-b
vif favor vif-a
vlan create vif-master 2 51 57
ifconfig vif-master-2 `hostname`-2 mtusize 1500 netmask 255.255.255.0 partner vif-master-2
ifconfig vif-master-51 `hostname`-51 mtusize 1500 netmask 255.255.255.0 partner vif-master-51
ifconfig vif-master-57 `hostname`-57 mtusize 1500 netmask 255.255.255.0 partner vif-master-57