ESXi 4.1 vs Hyper-V

[PigLover]

I have a conundrum...

I'm currently running Hyper-V with all windows-based clients on my server (Server08, WHS, etc). The server is built in a Norco 4220 case with all 20 drives populated, all running on a single HBA + expander. All of the clients run their OS on VHD, but data drives are passed through natively - most importantly, the 14 WHS drives get native passthrough to take advantage of WHS's NTFS compatible filesystem.

WHS is obviously a dead-end. Vail is effectively killed. I want to start experimenting with options, but all of the options I have are -nix based (freeBSD+ZFS, linux softraid, Nexenta, SolarisExpress, etc). And while Hyper-V plays nice with all of these OSs, the integration tools that let you use disk passthrough and higher-performance networking only work with Linux 2.4-based kernels that are quickly becoming obsolete. I've downloaded and played with the one 'officially supported' Hyper-V Linux integration - SUSE-enterprise server 10 - and it is just a PITA.

So I want to move to ESXi 4.1. I bought an LSI HBA to swap in from ESXi's (rather limited) compatibility list. All the rest of my hardware is compatible. But here's the conundrum: no supported method to do disk passthrough. Yes, you can pass through the whole HBA to a single guest OS, but that means all the disks on that HBA go to one guest OS. Not exactly what i had in mind. I'd like to leave the WHS passthrough in place and start testing other options on the renaming drives. And I'd ideally like to test them with passthrough too.

I've seen the instruction guides to set up 'unsupported' passthrough. Not real appealing...

Any suggestions? Best one I can come up with is run two servers - Hyper-V with Windows guests and ESXi for everything else. There have got to a better ideas out there.

 

[zpackrat]

keep in mind that you can pass through a single LUN carved out on the HBA to a guest. So in effect you could have 4 disks passed to 4 guests if needed. That's how I roll, 2 partitions on 1 raid set, 1 for VMFS and one passed through to my 08r2 file server. As for setting up the Raw Disk Mapping, it's easier than you might think.

 

[Child of Wonder]

keep in mind that you can pass through a single LUN carved out on the HBA to a guest. So in effect you could have 4 disks passed to 4 guests if needed. That's how I roll, 2 partitions on 1 raid set, 1 for VMFS and one passed through to my 08r2 file server. As for setting up the Raw Disk Mapping, it's easier than you might think.

This. Sounds like you need RDMs. However, bear in mind RDMs aren't necessarily going to provide any better performance than VMFS (except in rare or uncommon circumstances) and you lose portability and features available to that VM should you ever have multiple VMware hosts managed through vCenter.

 

[PigLover]

keep in mind that you can pass through a single LUN carved out on the HBA to a guest. So in effect you could have 4 disks passed to 4 guests if needed. That's how I roll, 2 partitions on 1 raid set, 1 for VMFS and one passed through to my 08r2 file server. As for setting up the Raw Disk Mapping, it's easier than you might think.

The problem with using LUNs carved out this way for WHS is this: you lose one of the key advantages of WHS if you put a VMFS wrapper around the drive. WHS drives are native NTFS and you can pull them from the drive pool and read them on any system that understands NTFS (which is pretty much any Windows or -nix system in the world). Doesn't work anymore if you put it inside VMFS.

RDMs are interesting and would probably solve the problem. Its no about performance, per se, but about keeping the entire drive 'native' to what the guest OS writes on it. But seriously...there's got to be a better way. At least there has to be somebody out there who has built scripts or a UI wrapper around doing this.

 

[zpackrat]

The problem with using LUNs carved out this way for WHS is this: you lose one of the key advantages of WHS if you put a VMFS wrapper around the drive. WHS drives are native NTFS and you can pull them from the drive pool and read them on any system that understands NTFS (which is pretty much any Windows or -nix system in the world). Doesn't work anymore if you put it inside VMFS.

RDMs are interesting and would probably solve the problem. Its no about performance, per se, but about keeping the entire drive 'native' to what the guest OS writes on it. But seriously...there's got to be a better way. At least there has to be somebody out there who has built scripts or a UI wrapper around doing this.

Keep in mind your working with what is considered to be an enterprise product, not going to be much support for home products. That said, using an RDM is the way to go. I can help you get them set up, it's easy as pie. as stated previously, I have a 2 4 disk raid 5 disks, one of which is VMFS, the other has 2 partitions carved out from the controller (perc 5i), one of which is a small VMFS store the other is presented as an RDM. My thinking is that I can plug the controller with drives into another system if ever needed to get to my files without the need for standing up another ESXi server.