FC Zoning in ESX 3.5 infrastructure with Brocade silkworm 200e's and IBM DS4700 SAN

[srenman1]

After Reading the fiber channel SAN config guide from vmware, (Chp 4 pg 58-62 on IBM DS4000 storage systems), I have a few questions. First the guide states that you should use MRU (most recently used) multipath policy however when taking with IBM support they recommend using LRU. Does anyone have any experience with this setting?

The guide also gives an example how to configure a highly available SAN failover design (pg 59). I have cabled mine the same as recommended in the guide. Attached is a picture of my situation. I am unsure if I am zonine this correctly. Ideally I would like to have this zoning setup in a "single initiator single target" (most secure) or single initiator multiple target configuration. There is only going to be (2) large VMFS luns grouped in a storage partitions in the storage manager that both the ESX servers need to see in order for HA and vmotion to work. Any tips to get me off in the right direction?

Inside the Storage manager I have created a Host group with a host type LNXCLVMWARE and made an Alias for each HBA (2 per ESX server) so that I can see the logical drives on the SAN. I have no zones setup in the switches right now and from the ESX boxes I can see the SAN storage disks and add them fine if I choose "add storage". SCSI Target 0 shows path to disk and when I click on manage paths, the first two show "dead" and then the last two are "on" and "active". By default do these switches to multipathing with out any config?

Anyone know any good tutorials on zoning practices for a vmware environment? I checked out the following, but need more info.

[AndreTheGiant]

Anyone know any good tutorials on zoning practices for a vmware environment? I checked out the following, but need more info.

In VMware environment zoning is quite simple: everybody must see everybody

Andre

**if you found this or any other answer useful please consider allocating points for helpful or correct answers

[caciolli · Accepted solution]

I have a lot of DS4700 in similar environments, zoning is easy.

Each host adapter must be zoned to two controller port, for example:

Host A HBA1 to CTRLA port A,

Host A HBA1 to CTRLB port B,

Host A HBA2 to CTRLB port A,

Host A HBA2 to CTRLA port B

Opposite ports for Host B

This is an example, someone prefer to use other combination (ABBA etc).

The result in this configuration is to see 4 paths to each LUN.

MRU is the only option for DS4000 family, this is default for ESX 3.5, just have a look to this paramenter.

Best practice reccomend to disable AVT for the host topology LNXCLVMWARE. I usually did not change this. In DS4000 firmware version 7.xx this parameter is disabled by default.

I also suggest you to remove the ACCESS LUN from the vmware host group.

[depping]

No it's not Andre. Hosts shouldn't be allowed to see each other. This causes unnecessary overhead when one of the hosts is rebooted, often referred to as Registered State Change Storms.

It's a best practice to do Single Initiator zoning. In other words you will need to create a zone for each iniator, in your case the HBA. The HBA will be zoned to both Storage Controllers. Of course you could even do it more rigid, single initiator / single target, but this is usually not needed. Anyway, for each HBA of an ESX host you will need to create a zone. To this zone you will need to add both Storage Processors. You should end up with 4 paths, and none of these paths should be dead.

In my opinion storage is one of the most important pieces of the VMware puzzle. If you don't know what you are doing or aren't certain you are doing the correct thing you should try to get someone in who helps you.

LRU ? On the VMware side you only have the option between MRU / Fixed. If VMware marks the array as MRU just let it be... Changing this setting can cause some of the weirdest problems you've ever seen.

Duncan

VMware Communities User Moderator | VCP | VCDX

-

---

Blogging:

Twitter:

If you find this information useful, please award points for "correct" or "helpful".

[AndreTheGiant]

No it's not Andre. Hosts shouldn't be allowed to see each other.

True. I had to rewrite in a right way: each "server" must see each "storage".

With some switches (like broadcom 200E) you can do this configuration automatically.

Andre

**if you found this or any other answer useful please consider allocating points for helpful or correct answers