ESXi Queries?

Soldato
Joined
18 Oct 2002
Posts
3,027
Location
Pentonville Prison
Some questions for the ESXi gurus. I've only used KVM/QEMU on Linux and under unRAID

1) Anyone done any passthrough of physical hardware like GPU, Soundcard? Do you need to stub or "blacklist" any hardware or is everything available for passthrough? Onboard USB controllers? Any issues with Nvidia cards (NV tries to block virtualisation generally) or with passing through a video card when it is the primary card on a system (I guess not since ESXi is happy to be headless)?

2) Can you specify CPU type (a useful tool in KVM is to be able to tell the VM that the CPU is xyz Intel model to allow an easier time when installing fussy VMs like OSX!) Or does it passthrough CPU cores "untouched" so AMD cores are identified as AMD (and OSX VMs would need hacked kernels!)

3) What about core/HT core pairings - ESXI seems to call all of them vCPUs. On LVM if you ran a machine on HT cores only it would run like a dog. How is that tackled?

4) The store for ISOs etc... Do the ISOs need to be there just for install? If so could I install ESXi on a 16Gb stick and use that for my ISO store as well and move ISO off after creating each VM?

5) Do VM HDDs support thin provisioning?

6) Can I import existing HDDs with existing data or must they all be wiped and converted to some VMware FS?

Thanks for any enlightenment you can offer ;-)
 
1) there are a number of people on hear with the N54L server which passthrough the GPU. my limited expirence suggest not passing thought the integrated onboard this seems to break esxi. if your passing though USB make sure your not booting esxi from usb.
2) not sure
3) does not seem to be a problem when i've tred it. typically i tend to only assign 2 cores per Vm, but when i assigne the whole 4 cores i dont notice any issues.
4) the ISO i belive need to be in the Datastores, can be removed once your finished with them
5) yes they support thin provisioning, and non-persistant disks so changes from spesific choosen time. also snapshots can be taken.
6) not directly as far as i know (unfermilliar with KVM/QEMU, but VMWARE have a Convertor you can install on the VM or remotely and conenct to the machine and V2V the machine into ESXI
 
6) not directly as far as i know (unfermilliar with KVM/QEMU, but VMWARE have a Convertor you can install on the VM or remotely and conenct to the machine and V2V the machine into ESXI

Thanks.. I mean if I had a disk with NTFS (I don't, mine are XFS) could I import it as it is - ie the hardware and pass that untouched through to a VM or must I go thru an intermediary file system
 
The datastores are formatted using VMFS, you can possibly mount the XFS partitions to read from but I wouldn't run VM's using it.

Best bet is to copy anything off, reformat using vmfs and upload it back to the datastore.
 
The datastores are formatted using VMFS, you can possibly mount the XFS partitions to read from but I wouldn't run VM's using it.

Best bet is to copy anything off, reformat using vmfs and upload it back to the datastore.

Sorry, I wasnt being clear. I want to keep my existing data (movies etc) and import my disks into a VM. I'd obviously format the datastore for storing the VM image/OS itself...
 
Sorry, I wasnt being clear. I want to keep my existing data (movies etc) and import my disks into a VM. I'd obviously format the datastore for storing the VM image/OS itself...

I'm sure you can do this. I remember having an NTFS formatted drive that i presented to the host so that a VM could access it as a second disk. I don't recall converting it to a vmdk, although this was many years ago so would need to check.

Edit: I believe it's called Raw Device Mapping (RDM), check this link out: https://www.experts-exchange.com/ar...-LUN-to-a-virtual-machine-hosted-on-ESXi.html

I find the ISO's quicker to run from when placed in a datastore, but they can be presented from the host via any drive or disk (cd/dvd).
 
Thanks.. I mean if I had a disk with NTFS (I don't, mine are XFS) could I import it as it is - ie the hardware and pass that untouched through to a VM or must I go thru an intermediary file system
Yes, it's called Raw Device Mapping.
 
4) The store for ISOs etc... Do the ISOs need to be there just for install? If so could I install ESXi on a 16Gb stick and use that for my ISO store as well and move ISO off after creating each VM?
The USB stick that you install ESXi onto is not a datastore (it's not formatted with VMFS), so you can't put ISOs (or anything else) on it.
 
MORE questions for those experienced....
7) Under KVM it makes a big difference for video or USB card passthrough whether you have devices in their own IOMMU groupings and sometimes you have to add an ACS patch to enable separations. Is there any such restriction for ESXi?
8) ESXi is headless and can be controlled by web UI so I can passthrough ALL VGA cards in system yes?
9) How do you determine which VCPUs are real cores and HT pairs? This could affect performance of VMs for games BIG time
10) Can you pass through ALL VCPUs? Are no resources retained for the ESXi admin overheads?
 
Last edited:
1) why are you trying to do this generally you dont need to?

2) not really that I have noticedpossibly by messing with hardware levels but the CPU im pretty sure identifies as what ever it is even if you set a low hardware comparability level (it jsut masks out new features)

3) ESXI knows what its doing you dont need to worry it will only run like a dog if you max out the CPUs

4) Store the ISOs on a vmfs datastore

5) Yes

6) Id use some imaging software to import the vms
 
7) dont know
8) Yes all our servers are headless
9) esxi does this... esxi and games are not two things id put together...
10) you can install 50 dual core vms on a quad core host if you want (we are over provisioned by about 500%)
 
7) dont know
8) Yes all our servers are headless
9) esxi does this... esxi and games are not two things id put together...
10) you can install 50 dual core vms on a quad core host if you want (we are over provisioned by about 500%)

Thanks... answer to (7) is I believe that ESXi does need IOMMU groupings... which is a problem on most of the new X399 boards and most X370, generally Intel boards work better it seems
 
9) How do you determine which VCPUs are real cores and HT pairs? This could affect performance of VMs for games BIG time
10) Can you pass through ALL VCPUs? Are no resources retained for the ESXi admin overheads?
You are coming at this from the wrong angle. There's no such thing as passing a VCPU through. The vCPUs that you present to a VM are *virtual*, and when the VM needs to execute something, the Hypervisor *schedules* that to be run on the *physical* CPU. How it does this is a big part of the secret sauce, and "not something you need to worry your pretty little head about", if that makes sense. The whole idea being that the VM will enjoy the maximum performance achievable given the current conditions. You can do things like assign shares to different VMs so that VM #1 gets priority over VM #2 whenever there is contention, but ultimately, the physical CPU is a shared resource.
 
Back
Top Bottom