19 Mar 2017
Introduction
What does that even mean?!
There are multiple names for the functionality I’ll be talking about today – PCI device assignment, device assignment, PCI passthrough, host device passthrough, hostdev passthrough… There really is no difference between them for the sake of this post. Let’s drop the fancy wording for a minute - we are talking about taking a device which happens to be on computer’s PCI bus and “passing it” to a virtual machine with ideally no performance loss.
PCI device within host operating system:
PCI device passthrough:
What can we achieve with that?
continue reading
23 Jan 2017
Introduction
KVM and QEMU support two paravirtualized storage backends: the older virtio-blk and more modern virtio-scsi. Fully emulated devices are not in scope of this post as their performance is subpar1 and shouldn’t be used except for compatibility purposes – like CD-ROMs. We won’t focus on the fine grained difference between virtio-blk and virtio-scsi. Why? You can find multiple benchmarks and comparisons online234. The idea is to go over the high-level difference between the backends as our main focus is to determine the suitability in enterprise deployments managed by oVirt.
continue reading
19 Sep 2016
Introduction
With VFIO and SR-IOV, virtualization is now capable of delivering network bandwidth and latency comparable to bare-metal machines. To exploit this in latency sensitive environment such as NFV, real-time (RT) operating system is required. oVirt does not really have an concept of RT hosts and guests at the moment, I was wondering whether it is possible to have such host nevertheless. As it turns out, with a bit of tweaking and hacking, it’s possible!
continue reading