Re: [Qemu-discuss] Best Intel hardware for qemu

[Jakob Bohm]

If you pass through the disk access to your SAN partitions as disk
accesses to block devices (such as SAN client drivers) in the host
machine, you don't need VFIO for that.  This can handle nearly
unlimited number of virtual machines without running out of PCI
slots in the host machine.  This is the equivalent of "passing
through a SAN disk" in VMWare, but isn't artificially limited to
SAN disks (for example, you can layer the Linux multipath drivers
and/or the Linux disk encryption drivers between the virtual
machine and the actual SAN).

If you pass through the network access to your (iSCSI or NBD) SAN
as network traffic via the general qemu/kvm network features, you
don't need VFIO for that.  This can handle nearly unlimited number
of virtual machines without running out of PCI slots in the host
machine.  This is equivalent to using a "VMWare virtual switch".

If you dedicate a physical SAN adapter (iSCSI, NBD, SAS or fibre
channel) to each virtual machine and pass it through to that
virtual machine, you need VFIO for that.  As on VMWare, this will
limit you to one virtual machine for each PCI slot in the
motherboard.

If you dedicate a physical network adapter (NIC) to each individual
virtual machine and pass it through to PCI drivers in that virtual
machine, you need VFIO for that.  This too will limit you to one
virtual machine for each PCI slot in the motherboard.

As for passing through raw SCSI devices or busses, I don't know
if the latest qemu versions have the ability to do this at a
hardware-independent level like VMWare does (VM sends standard
SCSI requests to qemu virtual SCSI adapter, qemu sends those
same SCSI requests to real SCSI hardware via something like the
Linux "SCSI generic" driver, optionally mapping at most the
SCSI-level bus address).

On 10/04/2019 19:42, Lars Bonnesen wrote:
> But for sure I want passthru - for running virtualized SAN and such
>
> Any unofficial list?
>
> Regards, Lars.
>
> On Wed, Apr 10, 2019, 18:58 Friedrich Oslage <address@hidden>
> wrote:
>
> > As long as it's VT-x capable and can run Linux, you're good to go.
> >
> > It only gets tricky once you start using VFIO (direct passthrough of
> > host PCI devices to a VM, such as GPUs, NICs or NVMes for instance). You
> > need VT-d support for that and both the CPU and mainboard have to
> > support it. And not only do they have to support it, they have to
> > support in a usable way with decent iommu group isolation and without
> > weird bugs. There are no (official) compatibility lists for this, it's
> > still mostly trial and error...
> >
> > Regards
> > Friedrich
> >
> > On 4/10/19 2:26 PM, Lars Bonnesen wrote:
> > > So I am coming from the VMware world (with comprehensive compatibillity
> > > lists) but about to start a project with KVM/Qemu and I would like to
> > setup
> > > an inexpensive test setup for this purpose.
> > >
> > > I am thinking of buying one of SuperMicros IoT-servers like
> > https://www.supermicro.com/products/system/Mini-ITX/SYS-E300-9D-4CN8TP.cfm
> > > Will this be a nice pick for Qemu? Or any VT-supportet system will work
> > > fine?
> > >
> > > Regards, Lars.


Enjoy

Jakob
--
Jakob Bohm, CIO, Partner, WiseMo A/S.  https://www.wisemo.com
Transformervej 29, 2860 Søborg, Denmark.  Direct +45 31 13 16 10
This public discussion message is non-binding and may contain errors.
WiseMo - Remote Service Management for PCs, Phones and Embedded