Ovirt devel - Dec 2009 - oVirtBIOS : (High-Performance) Virtualization Firmware

Dear Dennis,
> Why should I care? Don't get me wrong the idea sound interesting but I 
> don't really see why it is so vitally important to put the HV right 
> into the BIOS. The problem is that you loose support for a lot of 
> hardware that cannot be booted with coreboot.
Our engineering solution comes from a real world problem : faulty 
firmware (BIOS or UEFI) implementations prevent efficient I/O 
Virtualization. To quote David Woodhouse, lead Intel embedded software 
developer for the Linux Kernel and principal maintainer of the file 
drivers/pci/intel-iommu.c  [0] ( Intel's I/O Virtualization, to provide 
high performance inside the virtual machines ) [1] :

"Well done, Dell and HP -- although I didn't think it was possible, you
have _further_ lowered my already-unprintable opinion of closed source 
BIOSes and BIOS engineers"

"We _really_ need open source firmware.

Or at _least_ firmware written by competent engineers -- but I think
we've all fairly much given up on that happening by now?"

Many commits of the file intel-iommu.c are related to BIOS bugs [2] [3] 
[4] [5] [6] [7].

And it is evident that Intel will not be able to test each firmware 
version on each IOMMU capable Hardware.

By ensuring a full use of hardware capabilities inside Virtual Machines, 
through a custom Virtualization Firmware, it lowers the TCO (Total Cost 
of Ownership), thus requiring less frequent hardware upgrades that can 
be substantial saves on a large scale volume or a critical requirement 
for budget aware customers.
> Earlier this year we had to come up with a virtualization solution to 
> host 80 VMs quickly and our first shot was VMWare ESX but that failed 
> because ESX refused to work with the commodity hardware we were using. 
> So we went with RHEL Xen instead which works beautifully on pretty 
> much any system precisely because it isn't so closely wedded to any 
> particular hardware.
> I think using a regular BIOS that boots a minimal Kernel/Initrd from a 
> flash chip gives you pretty much the same benefits of a tiny footprint 
> but actually works with pretty much every machine out there.
We would greatly appreciate to invite you to a further reading of this 
Phoronix article entitled : "KVM Virtualization Performance With Linux 
2.6.31" [8] :

"The disk benchmarks took a large hit when running Ubuntu 9.10 "Karmic
Koala" within a virtual machine using KVM, but that is to no surprise 
considering the setup."

"Benchmarking Apache with the Kernel-based Virtual Machine took a huge 
performance hit,"


Our engineering target is to achieve 1:1 parity in terms of performance 
between real-hardware and a virtual machine.

Each product has specific needs that is why we are very pleased that you 
were satisfied by your Virtualization Solution, to quote : "So we went 
with RHEL Xen instead which works beautifully", however it will not be 
the case for performance and budget aware customers.

Best Regards,

Guillaume FORTAINE

[0] 
http://git.infradead.org/iommu-2.6.git/blob/e0fc7e0b4b5e69616f10a894ab9afff3c64be74e:/drivers/pci/intel-iommu.c
[1] http://patchwork.kernel.org/patch/42841/
[2] 
http://git.infradead.org/iommu-2.6.git/commit/e0fc7e0b4b5e69616f10a894ab9afff3c64be74e
[3] 
http://git.infradead.org/iommu-2.6.git/commit/2ff729f5445cc47d1910386c36e53fc6b1c5e47a
[4] 
http://git.infradead.org/iommu-2.6.git/commit/6ecbf01c7ce4c0f4c3bdfa0e64ac6258328fda6c
[5] 
http://git.infradead.org/iommu-2.6.git/commit/5854d9c8d18359b1fc2f23c0ef2d51dd53281bd6
[6] 
http://git.infradead.org/iommu-2.6.git/commit/86cf898e1d0fca245173980e3897580db38569a8
[7] 
http://git.infradead.org/iommu-2.6.git/commit/0815565adfe3f4c369110c57d8ffe83caefeed68
[8]
http://www.phoronix.com/scan.php?page=article&item=linux_2631_kvm&num=6

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://listman.redhat.com/archives/ovirt-devel/attachments/20091227/3e246b1b/attachment.htm>