KVM Forum 2019

Clever scheduling of virtual CPUs on Symmetric MultiThreaded systems for, among other things, making highly impractical side-channel attacks even more unpractical, is no new idea. Unfortunately, via exploiting L1TF and MDS vulnerabilities in Intel CPUs, impractical is becoming practical!

But, instead than disabling SMT, we can avoid that VM share cores. This is called core-scheduling, and implementing it requires quite some scheduler changes. Nevertheless, work toward that is being done for both KVM and Xen (and other hypervisors have it already).

After an overview of L1TF and MDS, we will see how core-scheduling may help and why it is so tricky to implement (although in different ways) for both KVM and Xen.

We will show numbers from performance evaluation of the currently available implementations. In fact, all this only matters if performance are better than turning SMT off.

RISC-V (pronounced "risk-five") is an Instruction Set Architecture (ISA) that's available under open, free and non-restrictive licences. It is a clean and modular ISA where new features are added as optional extensions. The RISC-V hypervisor extension provides virtualisation capabilities to a RISC-V CPU and it is designed considering both Type-1 and Type-2 hypervisors. In this talk Alistair and Anup will explain the RISC-V Hypervisor extension, discuss how it was implemented in QEMU and talk about the RISC-V implementation of KVM.

Alistair will talk about the current state of the RISC-V Hypervisor Extensions in QEMU. This will include details about the implementation and design choices that were made. He will discuss what we currently have upstream and how this compares to the latest and proposed future specification versions. This will include all known limitations and proposed future work in the QEMU implementation. He will also talk about current out of tree work that is not yet ready to be submitted upstream and discuss how this can be upstreamed.

Anup will then explain KVM RISC-V internals and the road ahead for KVM RISC-V. Anup will also show a demo of KVM RISC-V using using KVMTOOL.

VirtIO was designed to standardize hypervisor interfaces for virtual machines - but we are beginning to see the emergence of Virtio hardware. This talk will answer the questions: why does this make sense, what works and what are the issues hardware implementations of virtio have to overcome?
Topics to be covered:

- What is the difference between hardware virtio devices and virtio data path accelerators?
- What are the minimal requirements of virtio in hardware?
- How can we handle compatibility, including hardware bugs and limitations?
- How to make live migration work? What about overcommit?
- Which changes included in the recent virtio specification help design hardware virtio devices?
- Which known issues remain and how does the Virtio committee plat to address them?
- Why design Virtio in hardware? Are there alternatives?
- Why get involved with the Virtio specification process?

With the advent of fast storage technologies like NVMe and 3DXP, hypervisors are facing unprecedented challenges. The added software overhead involved in access validations, general data movement and notification between domains is more noticeable than ever. It affects all sorts of performance dimensions including bandwidth, IOPS and latency, most of which have been vastly hidden by the slow nature of devices until recently.

This talk is divided in two parts. Firstly, we will focus on storage performance evaluation and benchmarks, showing how these translate to virtualisation. Secondly, we will dive into hypervisors based on KVM and Xen to compare how they work and discuss how they can deliver the best end user experience in terms of performance and efficiency.

When it comes to performance monitoring KVM provides sophisticated tools to deep dive into specific aspects. For example kvm_stat or perf allow to analyse the exits from guest mode. On the other hand getting a system level view or having permanent monitoring and analytics is only available for the process view, e.g. with tools like sysstat. Other hypervisors offer a much better out of the box experience.
This talk is about extending the tooling for kvm stats to better integrate into the bigger picture.

A technical (and end-user oriented) Q&A panel discussion on a variety of topics related to KVM, QEMU and more. The discussion will be for about an hour. Topics will be chosen on the spot from a prepared list, and from the live Etherpad, where an audience (live or remote) can add questions before or during the discussion.

Etherpad: https://etherpad.net/p/KVMForum2019Panel