Connecting VM to iSCSI

[edtech]

Hi folks,

I have a ESX 3.5 infrastructure tied into some Equallogic storage.  Each host has two 1GB connections into the private iSCSI network.  I need to build some new fileservers with large storage allocations (up to 2TB).  I had debated about just making big vmdks for the fileservers or possibly RDMs but I'm looking more towards using software iSCSI on the VM and connect out to LUNs.

My question is, do I simply make a new vSwitch that is presentable to the VMs that utilizes the same vSwitch the host uses for its iSCSI connection?  The existing iSCSI vSwitch has a service console port and vmkernel port and is not visible to the VMs.  Is there any issue with both the host and guest using the same iSCSI connection? Would it be better to add additional NICs for the VMs to use for iSCSI?  Up to this point the VMs have all been using vmdks so I haven't crossed the bridge of needing a VM connect to iSCSI.

Is the iSCSI route not the best practice for this scenario, and should I reconsider vmdks/RDMs?

All suggestions are appreciated.  I'm hoping to migrate the infrastructure to 4.1 this summer if that has any impact on the recommendations.

Thanks!

[vmroyale · Accepted solution]

Hello and welcome to the forums.

My question is, do I simply make a new vSwitch that is presentable to the VMs that utilizes the same vSwitch the host uses for its iSCSI connection?

It depends.  If you have the NICs in each host, you could create a separate vSwitch and use this for guest iSCSI traffic.  I prefer this approach, when the resources are available.  There is some guidance on this separation in the "SAN best practices for deploying Microsoft Exchange with VMware vSphere" document.  With that being said, you can also just create a new virtual machine port group on your existing vSwitch and go that route.

The existing iSCSI vSwitch has a service console port and vmkernel port and is not visible to the VMs. Is there any issue with both the host and guest using the same iSCSI connection?

It might make things a bit more difficult to troubleshoot, but it is definitely a more simple setup to implement.

Would it be better to add additional NICs for the VMs to use for iSCSI? Up to this point the VMs have all been using vmdks so I haven't crossed the bridge of needing a VM connect to iSCSI.

More is always better right?  It really depends on the current (+ new) usage, but it probably wouldn't be a bad idea.

Is the iSCSI route not the best practice for this scenario, and should I reconsider vmdks/RDMs?

The 2TB-512byte limit is a good reason for considering the guest initiator.  This approach solves the 2TB-512byte limit, but it creates new complexities around VM portability, backup and other operational tasks.  Take these things into consideration as well.

Good Luck!

[AndreTheGiant]

If you have free pNIC you can build a vSwitch only for VM iSCSI connectivity.

But there is no problem to have the same of existing host iSCSI.

Andre

[edtech]

Thanks!

So what issues arise with the initiator path?

Backups currently are strictly file-based through backup agents (Backup Exec).  I have not yet deployed a solution to backup the whole VM but am exploring the options.

[vmroyale]

Your current backup approach will be unaffected, if you use the guest initiator.  This is a good thing.

The other issues are around portability, and include matching vSwitch VM port groups on each host (to ensure vMotion), and losing the ability to restore the VM as a single set of files.  Again, with the backup strategy you are using, you aren't really looking at many negatives with the guest iSCSI initiator approach.