Unsolved
This post is more than 5 years old
1 Message
0
91037
Why use a separate subnet for each iSCSI controller port MD3000i?
Can anybody give me an explanation of why I should use use a separate subnet for each iSCSI controller port with the MD3000i?
Its suggested as best practice in some of the documentation but not backed up with any reasoning, conversely I have also read that having all 4 controller ports in one subnet is a valid configuration if the switches are trunked/ISL'ed together.
I currently have a redundant (single) Cisco switch architecture (PSU's/Modules/Supervisors) and 1 storage VLAN with all the ports in one subnet. It works but I want to make sure I'm not missing a trick here and that failover/performance/load balancing might be better in a different config.
The documentation seems a bit light on actually explaining the pro's and con's of each setup.
Thanks,
Davey
Dev Mgr
9.3K Posts
0
March 30th, 2009 11:00
If you put all the ports on the same subnet, they need to be in the same vlan/switch.
In the rare case of a network storm or some other anomoly, your whole iSCSI network would go down.
If you have 2 NICs per server that you can dedicated to iSCSI I'd recommend at least 2 different iSCSI networks; each on it's own subnet and vlan/switch (if using vlans, make sure each vlan is on a different physical switch (or blade if you're using a big core switch), so that if a switch (or blade) dies you are still running on the other switch (blade).
The option for 4 different subnets/vlans is probably to give even more redundancy/robustness to your setup, though I've never seen that implemented.
mrokkam
154 Posts
0
March 30th, 2009 12:00
Davey,
this is what I got from some of the guys
"The recommendation is just to simplify the setup and configuration. The idea is that each port is connected to a different switch which in theory it is configured on a different subnet. Additionally, the recommended configuration would have two NICs on each server connected to each port on the controller. You would want those NICs on different subnets to facilitate the configuration on the server so it is clear what NIC is being used when you ping or you are establishing sessions. There is nothing wrong with the way you have it configured now."
The main problem is when you have multiple NICs in a server as shown below. If you don't setup the source ports of your iSCSI initiator and all NICs are on the same sub-net, only a single NIC would be used for all data as the stack would choose the first IP with the common subnet as the source IP. Also, if you don't have your switches connected/ISL'ed, you might be unable to ping/connect to some of the ports depending on the setup.
The best method, I believe, would be to have as many iSCSI subnets as you have host NICs in one server.
( image source: http://support.dell.com/support/edocs/systems/md3000i/en/2ndGen/IG/HTML/hardware.htm#wp998749 )
Also, do you remember which documentation you saw this recommendation in?
JOHNADCO
847 Posts
0
March 30th, 2009 15:00
Do a search on here with "a tale of two subnets" It's a good read.
One subnet isn't so bad, and can even be made just as redundant as two if done right with your switches.
Some actually prefer it this way.