Start a Conversation

Solved!

Go to Solution

1 Rookie

 • 

42 Posts

53

February 29th, 2024 17:08

ME4024: Cabling dual controllers to a VLT domain.

TLDR - The docs show cabling the controllers in a way that isolates each subnet to a switch. Why? I thought it would be ideal to have a path through each switch to both controllers via both subnets.

Obligatory "I'm not a storage guy" or "Network guy" for that matter.

We're trying to figure out the proper way to cable a dual controller ME4024 storage array to two S4048-ON switches that are configured for VLT (Dell's MLAG). We will be using all four CNC ports on each controller (8 total connections) and two subnets. iSCSI protocol to be used.

The following example from the official documentation has left us a little confused. If we're understanding this correctly, the docs are telling us to cable the CNC ports to the switches in such a way that one subnet is on Switch-A and the other subnet is on Switch-B. Highlights have been added to show subnets 10 (green) and 11 (blue). NOTE: The controller ports are 3, 2, 1, 0 (left to right) btw.


This doesn't make much sense to us. With this configuration I would think that Windows iSCSI initiator would lose connection to some targets...therefor losing connection to some storage. Unless I'm wrong and they won't actually lose access to storage...they'll just lose a path or two.

  • Wouldn't you want a path through each switch for each controller AND subnet?
  • If a switch is lost in the above example, access to an entire iSCSI subnet is also lost, no?
  • Is there a reason for isolating the iSCSI subnets to their own switches?
  • Are there different cabling recommendations if the switches are configured in VLT?

Update: Maybe my confusion here is with the subnet. I just noticed that the IP addresses are 192.68.xx.xxx. Am I looking at one 255.255.0.0 subnet in the example then? I wouldn't think so since the table clearly lists 10 and 11 in the subnet column as if they are two distinct subnets. I'm so confused by Dell's documentation.

Moderator

 • 

8.5K Posts

March 1st, 2024 14:26

The_LostIT_Guy,
 
To me it looks like that should indeed work, just make sure to vlan the traffic,, as Support Matrix has rules for switch settings. You would want to keep it simple, 1 subnet per switch, no trunking or lagging the iSCSI traffic, and each host should have 2 nics, 1 for each subnet
 
Let me know if this helps.
 
 

(edited)

1 Rookie

 • 

42 Posts

March 1st, 2024 13:52

Adding some more thoughts on this...

Based on the subnet/address to port mapping in the example above, we thought something like the following would be appropriate:

  • CNC ports 0,1 to one switch.
  • CNC ports 2,3 to the other switch. This makes each subnet available in both paths (Switch A and Switch B).

OR

Since each pair of ports (0,1 & 2,3) on each controller share a CNC chip, make ports 0/1 subnet 10 and ports 2/3 subnet 11 . . . cable as shown above.

1 Rookie

 • 

42 Posts

March 1st, 2024 15:46

@DELL-Chris H​ Thanks for your response. I'm assuming you're talking about the cabling and IP/subnet example from the Dell docs. 

- One subnet per switch, VLANs, no trunking/lagging iSCSI - Understood ✔.

- The hosts will have two NICs, one for each subnet. 

There should be no issues with using VLT? Would there be any scenarios where iSCSI traffic may be routed over the VLTi and if so, would that be a problem? 

 

(edited)

Moderator

 • 

8.5K Posts

March 1st, 2024 16:29

There shouldn't be any issue with VLT, and as far as the iscsi question, you would not want any routing of the iSCSI traffic.

 

1 Rookie

 • 

42 Posts

March 1st, 2024 16:36

@DELL-Chris H​ "Routed" probably wasn't the correct word to use.

Would there be any scenarios where the iSCSI traffic of one switch would traverse the interconnect to the other switch and if so, is that problematic?

(edited)

Moderator

 • 

8.5K Posts

March 1st, 2024 16:44

It should not be allowed to, it should be host to switch to port. Now you would be able to configure it to do so, but we do not suggest it.

 

1 Rookie

 • 

42 Posts

March 1st, 2024 18:22

@DELL-Chris H​ I assume this is because VLTi ports are not automatically added to VLANs that are not present in both VLT peers. I know for sure this is a feature of OS10 (I'm looking at some docs), but is it the same with OS9...is the VLTi only added to VLANs that are present on both switches? I would think it is. 

(edited)

Moderator

 • 

8.5K Posts

March 1st, 2024 21:20

The_LostIT_Guy,
 
Unfortunatly I don't believe it is included in OS9. 
 
 

1 Rookie

 • 

42 Posts

March 1st, 2024 22:01

@DELL-Chris H​ Hmm. Does this mean that . . . 

- ALL VLANs are automatically added to the VLTi in OS9?

OR

- NO VLANs are automatically added to the VLTi OS9?

Anyway, if it ends up being that all the VLANs end up on the VLTi in OS9...is UFD a sufficient method of blocking the traffic from one subnet over a switch? I would think that would work. Kind of a hack.

(edited)

Moderator

 • 

3.1K Posts

March 4th, 2024 02:06

Hi,

 

All vLAN are automatically added to the VLTi in OS9.

 

I can't be sure UFD would be the best way to block traffic from one subnet. If you manage to find out if UFD can, do update us and share your findings. 

1 Rookie

 • 

42 Posts

March 4th, 2024 13:13

@DELL-Joey C​ Will do. Thanks again for all the info. 

No Events found!

Top