[DanielJay]

I am in the process of setting up 2 new Dell PowerConnect 6224 switches for a redundant iSCSI network. My current setup has a Dell MD3000i with all 4 iSCSI network connections going into the same network switch. Then 1 NIC in each of our VMware ESX servers is connected to this iSCSI switch providing the connection to the storage. This is not the most desired setup due to the fact that we have a single point of failure - the network switch. What I am looking to do is setup 2 connections to each ESX server for iSCSI traffic with each link being connected to a separate switch. Then having the 4 iSCSI connections on the back of the MD3000i SAN be split between the 2 network switches. This way if one switch dies we do not loose all connectivity to the SAN. I have talked with Dell and I have been told to enable portfast and flowcontrol on both switches. Will I need to setup a LACP LAG between the 2 network switches or do not connect the 2 switches together at all?

Can anybody point me towards some articles in which I can learn more about setting up redundant high availability networks?

[zx10guy]

You won't need to link the two switches together in with a LAG connection. Just make sure you matrix your connections between the multiple NICs on your VM servers and the dual controllers on the MD3000i across the two switches. You want to make sure that in addition to one switch going out that you also account for a controller failure on the MD3000i.

As far as articles go, here's one straight from Dell: http://www.dell.com/downloads/global...ractice-WP.pdf

Page 5 of that article shows exactly what I'm talking about in my first paragraph of my response.

[DanielJay]

Dell PowerConnect department ended up sending me some documents after I talked with them yesterday. This first document deals with a little bit different SAN than we have but they stated the principles should still apply. In this document they state that non-stackable switches should be connected together. This information is stated on pages 10-12.
http://docs.danielkassner.com/dell/I...zation_EQL.pdf

Then in this second document on page 3 the document shows both switches connected together.
http://docs.danielkassner.com/dell/m...ploy_guide.pdf

When I contacted the Dell storage department for questions related to the MD3000i the support rep had no clue what Link Aggregation (LAG) was.
I am still reading these documents trying to determine the best setup.

[zx10guy]

LOL.

Sorry. But I've been down this path with Dell. The first document you linked to are some of the iSCSI SANs I'm using currently in my environment. Not my choice and have had a very heated battle with the individual who recommended them and brought them into the environment.

You don't need to have the switches linked together to your MD3000i. Even in the installation guide for the MD3000i, they don't call for the two switches to be linked together. http://support.dell.com/support/edoc...DF/IGbk02g.pdf Look on page 20. The reason why the second document you've linked to says to link the two switches together is because they are recommending a teaming setup. This would require the NICs to be able to see each other across the fabric. Depending on your teaming arrangement, you may be able to get away with just creating a LAG between the switches. If you are using LACP and doing a full up Etherchannel/port channel with your server NICs, you MUST use switches that can be stacked to look like one big virtual switch.

The reason why the first document you've linked to requires stacked switches is because of how Equallogic iSCSI SANs work. Even though each interface on the controller requires an IP address, there is a VIP that floats across all the ports which you use as your target for your initiators.

It's no surprise that the Dell guys have no clue about LAGs/Etherchannels/port channels. The reason I had a LOL moment was because I got into a heated debate with one of their pre-sales engineers on their Equallogic boxes. LAGs/Etherchannels/port channels are an unsupported configuration when used with their Equallogic product. I asked them why and never got a valid explanation. I got explanations that I was over subscribing the connection coming in to the SAN. Uhhhhhh.....NO. The pre sales engineer tried to convince me that I was feeding a full 2, 3, 4 (whatever the number of single Gig connections I used) Gb connection into a single 1 Gb connection on their controller. Anyone who understands networking and how Etherchannels work, knows that this is not the case. Then he tried to tell me I would be screwing up how their MPIO driver worked. I then told him I tested the configuration both without and with an Etherchannel set up on my server. And guess what? I got the same number initiator connections and the same measured throughput/performance. The pre sales rep had to step in to the argument to break it up between us.

With your MD3000i, you don't have a floating VIP. And to manage the unit, you have a management port which shouldn't be part of the iSCSI network fabric anyways. So you can set up a LAG between switches if you want to. But it is not necessary per what I've said.

Hope this clears it up for you.

[zx10guy]

I also forgot to add some other reasons why the switches need to be stacked for the Equallogic boxes.

Equallogic works on a virtualization model. When you add additional trays (members) to the storage group, the devices talk to each other over the same iSCSI front end, your MD3000i talks to additional trays via the expansion bus on the controller. This is one of my gripes about the Equallogic boxes. In addition to the overhead chatter that the controllers transmit to each other, the Equallogic controllers have this mystical algorithm which I have not been privi'd to. This algorithm shuffles around the location of block data from one storage unit to another based on performance metrics it gathers which is also a mystery to me. The only feedback I got is that it's proprietary and their secret sauce. And that I shouldn't worry about what it's doing as this is why their unit is so attractive. I hope you see something wrong with this picture.

[DanielJay]

I understand your pain zx10guy!

With my experience with stacking on the Dell PowerConnect 6224 switches is that if the master switch goes down than the stack reboots and re-assigns a master switch. During that time the machines would lose connectivity which would defeat the purpose of having a redundant network.

I did some more searching on the Dell Community forums and found this and this article. It appears that it is more common for people to setup a LAG between the switches after the SAN and servers (our ESX servers) have been connected properly to each switch.

Since I am not the original person that configured/setup the SAN/ESX servers I did some looking around. All 4 NICs on our SAN are in the same subnet and configured with IPs of 192.168.1.20-23. Then our ESX servers as they are configured now our iSCSI connections are teamed. One of the 2 connections is on a Broadcom based hardware, the other is an Intel Pro 1000 card. Now from previous reading I have come to understand that NICs cannot be teamed across multiple switches, however after reading this Q/A page it states that the Intel NICs should be able to be connected to multiple switches while teamed, but with only 1 active link and the other in standby mode.

Looks like I may have to look at the proper configuration for the ESX servers for what we are trying to do.