Virtual Bus

Virtual Bus

[Tej]

Are xen ppl are thinking to leverage Virtual Bus Technology recently
discussed on LKML
http://lkml.org/lkml/2009/3/31/339
http://developer.novell.com/wiki/index.php/Virtual-bus


thanks
-tej

Re: Virtual Bus

[Keir Fraser]

On 06/04/2009 16:09, "Tej" <bewith.tej@gmail.com> wrote:

> Are xen ppl are thinking to leverage Virtual Bus Technology recently
> discussed on LKML
> http://lkml.org/lkml/2009/3/31/339
> http://developer.novell.com/wiki/index.php/Virtual-bus

We probably could, with some small Xen-specific additions to the Virtual Bus
framework. However it'll be interesting to see whether it's a useful design
point between our fully PV drivers (which have pretty decent performance)
and full device passthrough (where there's no need to go through dom0 at all
on the data path). In particular, we can do device hotplug with device
passthrough, which may give us the flexibility that VB promises anyway.

 -- Keir

Re: Virtual Bus

[Espen Skoglund]

[Keir Fraser]
> On 06/04/2009 16:09, "Tej" <bewith.tej@gmail.com> wrote:
>> Are xen ppl are thinking to leverage Virtual Bus Technology recently
>> discussed on LKML
>> http://lkml.org/lkml/2009/3/31/339
>> http://developer.novell.com/wiki/index.php/Virtual-bus

> We probably could, with some small Xen-specific additions to the
> Virtual Bus framework. However it'll be interesting to see whether
> it's a useful design point between our fully PV drivers (which have
> pretty decent performance) and full device passthrough (where
> there's no need to go through dom0 at all on the data path). In
> particular, we can do device hotplug with device passthrough, which
> may give us the flexibility that VB promises anyway.

One can also instantiate a virtual PCI bus in dom0 populated with
virtual devices (which can be mapped down to real devices).  As far as
I understand vbus, this fulfills all the promises of vbus without
introducing a new bus concept.  It allows you to use legacy PCI device
discovery mechanisms and even enables mapping the virtual devices to
other guests (as PCI passthrough).

	eSk

Re: Virtual Bus

[Daniel P. Berrange]

On Mon, Apr 06, 2009 at 08:39:03PM +0530, Tej wrote:
> Are xen ppl are thinking to leverage Virtual Bus Technology recently
> discussed on LKML
> http://lkml.org/lkml/2009/3/31/339
> http://developer.novell.com/wiki/index.php/Virtual-bus

As I was reading that thread, the discussion seemed to descend into a
debate about whether this could all be achieved just by optimizing
the existing VirtIO driver backends / host, or whether the Virtual-bus
concepts could be incorporated into VirtIO in some way. So its probably
a little premature to talk about integrating this in Xen. Now an
interesting project that could be tried today is to setup Xend/QEMU-dm
to be able to use VirtIO drivers and then do some performance comparisons
between VirtIO & netfront/back & blkfront/back. The backends for VirtIO
are all implemented entirely within QEMU, so in theory it ought to be
easy to get them working under Xen.

Daniel
-- 
|: Red Hat, Engineering, London   -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org  -o-  http://virt-manager.org  -o-  http://ovirt.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-  F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

Re: Virtual Bus

[Gerd Hoffmann]

 > Now an
> interesting project that could be tried today is to setup Xend/QEMU-dm
> to be able to use VirtIO drivers and then do some performance comparisons
> between VirtIO&  netfront/back&  blkfront/back. The backends for VirtIO
> are all implemented entirely within QEMU, so in theory it ought to be
> easy to get them working under Xen.

Doesn't work.  virtio in the current qemu stable branch (which is the 
base for latest qemu-dm) is too old.  It doesn't use the dma mapping api 
(yet), which is required for it to work in qemu-dm.  Also the qemu block 
layer gets some serious rework right now, so the test results would be 
obsolete quickly.  I'd wait with such experiments until qemu 0.11 is 
released and merged into qemu-dm.

cheers,
   Gerd