KVM Forum 2014: Full Schedule

9:00am CEST

Keynote: KVM - Paolo Bonzini

Speakers

Paolo Bonzini

Red Hat

I've been working since 2009 on virtualization for Red Hat, where I am a Principal Software Engineer. My contributions have focused almost exclusively on QEMU and KVM since 2011, and since May 2013 I have been co-maintaining the KVM hypervisor. I've presented my work on QEMU at KVM... Read More →

Tuesday October 14, 2014 9:00am - 9:15am CEST
Room 3

Keynote KVM

9:15am CEST

Security Hardening of KVM - Andrew Honig, Google

This talk will cover a project to create a safer mode for KVM. We first started with an analysis of the guest reachable attack surface and historic vulnerabilities. Many of the vulnerabilities have been found in code that could be moved into userspace without significant performance impact to modern guests. The goal of the project is to provide a more flexible userspace API to allow userspace to handle more types of requests, while still leaving performance critical code in the kernel.

Speakers

Andrew Honig

Google

Andrew Honig is a tech lead on the Cloud Security Team at Google responsible for the security of Google’s Cloud Platform. He has discovered critical security vulnerabilities in KVM and VMWare virtualization products. He teaches courses in reverse engineering and software vulnerabilities... Read More →

Tuesday October 14, 2014 9:15am - 10:00am CEST
Room 3

KVM

10:00am CEST

Validating and Defending QEMU TCG Targets - Alex Bennée, Linaro

With ARMv8 silicon still scarce there is demand for a QEMU based solution to replace ARM's non-free Fast Models. We describe the challenges in providing such a port along with some novel techniques used to verify the implementation. We used a tool called RISU, which creates pseudo-random sequences of instructions which can be run in lock-step between a master and slave to verify emulation against a known good reference. We will discuss the limitations of the tool and analyse the bugs that were discovered after the initial patches were merged upstream. We will compare the results of code coverage analysis with the traditional black box testing used to validate system emulation. Finally we shall discuss how the testing could be better integrated into the project and how the lessons learnt could be applied to other architectures.

Speakers

Alex Bennée

Linaro

Alex is a senior software engineer working in Linaro's Virtualization team. An experienced FLOSS developer with over 20 years of experience in embedded and systems programming he currently spends most of his time on QEMU's TCG based emulation. The first piece of assembly he wrote... Read More →

Tuesday October 14, 2014 10:00am - 10:45am CEST
Room 3

QEMU

10:45am CEST

Break

Tuesday October 14, 2014 10:45am - 11:15am CEST
Foyer

11:15am CEST

ARM KVM Platform Device Assignment - Eric Auger, Linaro

The VFIO platform driver enables user-side accesses to platform device (non-PCI) resources. We leverage this work in a new QEMU VFIO platform device. A key challenge is poor IRQ performance. It turns out that porting on ARM existing legacy PCI frameworks based on eventfds, kvm irqfds and VFIO virqfds is not sufficient to reach near-native performance. To reach best performance, we propose to enable ARM GIC features that were not yet enabled in the kernel: dual stage priority drop and direct physical IRQ completion. Integrating these features bring challenges that require us to re-consider both the usage model and implementation of the VFIO platform driver and IRQFD framework. We will present the various IRQ assignment techniques and provide performance numbers based on Calxeda Midway xgmac assignment. We will discuss the implementation and integration options.

Speakers

Eric Auger

Linaro

Eric is an STMicroelectronics Linaro assignee, working in the virtualization team since Feb 2014. His work focuses on KVM platform device passthrough. He has previously worked at STMicroelectronics/STEricsson in various positions, latterly as a system architect assigned to iommu introduction... Read More →

Tuesday October 14, 2014 11:15am - 12:00pm CEST
Room 3

Network Functions Virtualization/Passthru

12:00pm CEST

Extending KVM Models Toward High-Performance NFV - Jun Nakajima, Intel Open Source Technology Center

As we have more demanding requirements for NFV (Network Functions Virtualization), such as handling smaller packets at very high rate in a scalable fashion, efficient and fast network-packets handling in middle boxes are critical. The architecture and performance of the virtualization I/O technologies, such as virtio rings, shared memory, and SR-IOV, can be improved by having a trusted entity that understands solution-specific data structures and that has access to guest memory. By trusting part of the guest and providing it with a memory view protected from the rest of the guest, we have extended the KVM models. We present 1) overview of the EPTP switching feature, 2) how we can provide protected memory views to trusted components in guests, 3) how we can improve and extend the current KVM models toward NFV combined with SR-IOV, and 4) best practices.

Speakers

Jun Nakajima

Intel Open Source Tecnhology Center

Jun Nakajima is a Principal Engineer leading open source virtualization and cloud projects, such as, KVM, Xen, and OpenStack at the Intel Open Source Technology Center. Jun has been working on various virtualization projects for almost a decade, and NFV is one of his ongoing projects... Read More →

Tuesday October 14, 2014 12:00pm - 12:45pm CEST
Room 3

Network Functions Virtualization/Passthru

12:45pm CEST

Lunch

Tuesday October 14, 2014 12:45pm - 2:00pm CEST
Foyer

2:00pm CEST

Optimizing IO Virtualization and VM Memory Management for Tablet Devices - Bokdeuk Jeong, Samsung

In this talk, we present our production-level optimization effort in both Android and QEMU/KVM for bringing out the virtualization tech to tablet devices. There are two key requirements for running Windows as an Android app: 1.To achieve good performance in terms of user perception; Starting up a Windows VM and switching from and to other apps should be done in a short period of time. Human interactive IO devices should be in no delay. 2.To seamlessly interface QEMU with Android framework. We develop automatic ballooning with low memory killer and activity manager. We also implement virtual IO devices such as multi-touch, audio w/ openSL, bluetooth w/ Bluedroid, etc. In addition, we optimize VM snapshot loader by modifying qcow2 and utilizing transparent huge pages. As a result, VM boot-time is reduced by 90% and VM performance is not compromised with IO delay on a recent Atom tablet.

Speakers

Bokdeuk Jeong

Samsung Electronics

Bokdeuk Jeong is a senior engineer at Software R&D center in Samsung Electronics. She is currenlty involved in a mobile virtualization project.

Tuesday October 14, 2014 2:00pm - 2:30pm CEST
Room 6

Kernel

2:00pm CEST

Hardware Accelerated Virtio Networking for NFV - Varun Sethi & Yashpal Dutta, Freescale Semiconductor

Network function virtualization (NFV) leverage's virtualization to consolidate multiple network functions on to a standard platform.The network functions are decoupled from hardware, allowing them to run in software as virtual machines. Network I/O with in these virtual machine needs to be both fast and flexible. Virtio networking offers a flexible mechanism for sharing I/O interface among virtual machines. Despite the flexibility, there are performance challenges with virtio/ vhost-net for virtual network functions. For high traffic rates efficient packet classification and distribution is required with virtio for VMs. This can be achieved using hardware offloads. In this presentation we discuss the performance issues encountered and how these were mitigated with hardware offloads while running IP forwarding as a virtual network function under KVM hypervisor with Vhost-net networking.

Speakers

Yashpal Dutta

Yashpal Dutta is a Lead Design Engineer at Freescale Semiconductor India. He is involved in enabling network and security acceleration on Freescale SOCs. He has expertise in the performance analysis and optimizations for network and security applications. He has also worked on performance... Read More →

Varun Sethi

Freescale Semiconductor

Varun Sethi is a Software Architect at Freescale Semiconductor and has been involved in virtualization software development for embedded Power Architecture SOCs. He has contributed to the KVM port for BookE.HV platforms and the e500mc core. He is the maintainer for Freescale PAMU... Read More →

Tuesday October 14, 2014 2:00pm - 2:30pm CEST
Room 3

Network Functions Virtualization/Passthru

2:30pm CEST

Automatic NUMA Balancing - Rik van Riel, Red Hat

In NUMA systems, each CPU has its own bank of memory, resulting in fast access to local memory, and slower access to memory elsewhere in the system. Recently a mechanism has been implemented in the Linux kernel to automatically run programs near their memory, and to move memory to near the programs using it. This presentation explains why computers are built this way, why NUMA locality matters, how the automatic NUMA balancing kernel code works, what it can do, and what kind of performance improvements have been observed. This presentation is also a good opportunity to discuss recent and future developments for the automatic NUMA balancing code.

Speakers

Rik van Riel

Red Hat

Rik van Riel is a principal software engineer at Red Hat, and a long term contributor to the Linux kernel. He has contributed to the memory management subsystem, the scheduler, and various components related to virtualization. Rik is active in community projects like kernelnewbies.org... Read More →

Tuesday October 14, 2014 2:30pm - 3:00pm CEST
Room 6

Kernel

2:30pm CEST

vhost-scsi Technical Discussion - Nicholas A. Bellinger, Datera

The discussion is around KVM disk I/O performance + latency using different approaches including vhost-scsi + in-kernel target, QEMU virtio-scsi, and virtio-blk/data-plane. This includes current developments for scsi-mq, and benefits that a blk-mq based I/O stack brings to the virtio-scsi LLD. It also covers remaining TODOs for supporting live migration in vhost-scsi code, and extra considerations with in-kernel LIO target state like SPC-3 persistent reservations and ALUA port access state across multiple nodes utilizing shared storage. Other topics include vhost-scsi support in Openstack, guest support for T10 Protection Information (PI), DIF metadata for application tagging of individual in-guest storage blocks, and different use cases for vhost-scsi vs. virtio-scsi dataplane + user-space iSCSI initiator.

Speakers

Nicholas A. Bellinger

Datera

I'm the SCSI (LIO) target maintainer for the Linux kernel, and co-founder + CTO at Datera. I'm also the principle author of the upstream iscsi-target, iSCSI extentions for RDMA (iSER) target, qlogic 4/8/16 Fibre Channel target, and vhost-scsi drivers. My ongoing projects include vhost-scsi... Read More →

Tuesday October 14, 2014 2:30pm - 3:00pm CEST
Room 3

Storage

3:00pm CEST

Break

Tuesday October 14, 2014 3:00pm - 3:30pm CEST
Foyer

3:30pm CEST

KvmGT: A Full GPU Virtualization Solution for KVM - Jike Song, Intel

GPU virtualization is an enabling technology in emerging virtualization scenarios. Unfortunately, existing GPU virtualization approaches are still suboptimal in performance and full feature support. iGVT-g is a product level GPU virtualization implementation with: 1) full GPU virtualization running native graphics driver in guest, and 2) mediated pass-through that achieves both good performance and scalability, and also secure isolation among guests. In this session we will present our work to enable iGVT-g in KVM. With this work, multiple KVM guests could all run 3D workloads with close-to-native performance. It's observed that there will be remarkable improvement of graphics performance for KVM Guest.

Speakers

Jike Song

Intel Corporation

Jike is working for Intel Corporation, focused on Graphics Virtualization - a full GPU virtualization solution Mediated Pass-Through. Previously this project was called XenGT, with only Xen hypervisor supported. Now it has a new name iGVT-g, and a KVM prototype, which will be referred... Read More →

Tuesday October 14, 2014 3:30pm - 4:00pm CEST
Room 6

Graphics

3:30pm CEST

KVM on MIPS - James Hogan, Imagination Technologies

The MIPS architecture is used across the spectrum from heavyweight networking gear down to wearables and embedded/IoT devices, which makes virtualization important on MIPS both for traditional use on servers and also for security on consumer hardware. In this presentation, James Hogan will look at the different approaches to virtualizing the MIPS architecture with and without hardware assistance, and will delve into some of the technical considerations involved in virtualizing MIPS with KVM.

Speakers

James Hogan

Imagination Technologies

James Hogan works in the MIPS processor division of Imagination Technologies. Prior to Imagination's acquisition of MIPS in 2013 he worked on a port of Linux to Imagination's Meta processor architecture, including upstreaming and now maintaining it, and also implemented a QEMU frontend... Read More →

Tuesday October 14, 2014 3:30pm - 4:00pm CEST
Room 3

KVM

4:00pm CEST

VFIO, OVMF, GPU, and You - Alex Williamson, Red Hat

Last year we introduced VFIO-VGA support, opening up virtual machines to be used as gaming platforms and high performance visual workstations. While many users have had success with this, we continue to struggle with the fragile nature of VGA routing and arbitration of this shared resource between host devices. In this talk we'll discuss some of these issues and present an alternative, legacy-free approach that avoids the VGA address space.

Speakers

Alex Williamson

Red Hat

Alex Williamson has been contributing to the Linux kernel and other open source projects for over ten years and is the maintainer of the VFIO userspace driver interface in the Linux kernel and VFIO-based PCI device assignment in QEMU. Alex currently works for Red Hat from his home... Read More →

Tuesday October 14, 2014 4:00pm - 4:30pm CEST
Room 6

Graphics

4:00pm CEST

KVM on IBM POWER8 Machines - Paul Mackerras, IBM Linux Technology Center

Recently IBM has started shipping POWER8-based servers with a KVM-based hypervisor. This hypervisor, IBM PowerKVM, is based on the work we have been doing for the last 4 years to make KVM run on IBM Power processors. Users can create and manage guests using the traditional virsh command line or via the Kimchi web interface, or can install OpenStack agents and manage the system using OpenStack. This talk will discuss the details of the IBM PowerKVM implementation, and look at the features and characteristics of the IBM POWER8 processor that relate to the task of making KVM run efficiently on the processor. This includes the new "micro-threading" mode which allows us to have up to 4 guests running simultaneously on a single core at any given point in time, but which introduces substantial implementation complexity in the architecture-specific KVM code to manage it.

Speakers

Paul Mackerras

IBM's Linux Technology Center

Paul Mackerras works in IBM's Linux Technology Center. He has been contributing to the Linux kernel for 18 years, and was the PowerPC Linux maintainer for 7. For the last 4 years, he has been working on making KVM exploit the hardware virtualization features of IBM's Power microprocessors... Read More →

Tuesday October 14, 2014 4:00pm - 4:30pm CEST
Room 3

KVM

4:30pm CEST

Graphics in QEMU -- How the Guest Display Shows Up in Your Desktop Window - Gerd Hoffmann, Red Hat

This talk shows how the graphics workflow works in qemu, to make the pixels the guest places into the cirrus vga memory appear in your gtk window or vnc client. We'll have a sneak preview into the future with upcoming opengl support and virtio-gpu emulation, bringing accelerated 3d graphics to QEMU. Recent changes such as multiseat support added to QEMU 2.1 will be covered too.

Speakers

Gerd Hoffmann

Red Hat

Gerd Hoffmann is working on virtualization. He started a few years back with user mode linux. Later the focus shifted to Xen. Nowdays he is working on qemu and kvm for the Red Hat. Currently he maintains spice and usb subsystems in qemu. Gerd gave various talks on virtualization-focuced... Read More →

Tuesday October 14, 2014 4:30pm - 5:00pm CEST
Room 6

Graphics

4:30pm CEST

VFIO on POWER - Alexey Kardashevskiy

VFIO is a virtual fabric IO mechanism which provides access for the userspace to real devices (now PCI only). One of possible uses is passing PCI devices to guests via virtualization software such as QEMU. In this presentation, Alexey will talk about differences between different IOMMU models, challenges it causes on POWERPC which include architecture differences, performance issues and interaction with KVM. Alexey is going to discuss with the community what could have been done different or better.in

Speakers

Alexey Kardashevskiy

2010-present - IBM Australia, LTC, Ozlabs team since 2010 on KVM-on-POWER, specifically on QEMU (sPAPR machine support) and VFIO (adapt originally developed for x86 driver to work on POWER7/8); 2007-2010 - IBM Russia, STG, developed SAS/Ethernet drivers for CellBE blade system firmware... Read More →

Tuesday October 14, 2014 4:30pm - 5:00pm CEST
Room 3

KVM

5:00pm CEST

QEMU Summit (Invitation-Only)

Tuesday October 14, 2014 5:00pm - 7:00pm CEST
Room 3

QEMU

9:00am CEST

Keynote: QEMU - Andreas Färber

Speakers

Andreas Färber

Wednesday October 15, 2014 9:00am - 9:15am CEST
Room 3

Keynote QEMU

9:15am CEST

New to QEMU: A Developer's Guide to Contributing - Jeff Cody, Red Hat

QEMU is a large, and active, open source software project, with many contributors. Some contributors have been involved in the community for a long time, while others are new to QEMU (and perhaps new to open source development, as well). For those 'newbies', it can be a daunting task to create properly formatted patches, interact with the community, and become a productive QEMU developer. This talk will focus on getting new potential QEMU developers through the hurdles of submitting their first patch, and what to expect afterwards.

Speakers

Jeff Cody

Red Hat

Jeff is a Senior Software Engineer with Red Hat, currently working in the virtualization group. He currently works on the QEMU block layer, with a focus on live block job operations, and block image formats. Jeff was a presenter at the 2012 and 2013 KVM Forums. He works remotely from... Read More →

Wednesday October 15, 2014 9:15am - 10:00am CEST
Room 3

QEMU

10:00am CEST

Nested Virtualization - State of the Art and Future Directions - Jan Kiszka, Siemens; Bandan Das, Red Hat; and Yang Zhang, Intel

Nested Virtualization was once considered as a research project, but we are seeing real demands for it, even in the cloud. Without nested virtualization, hardware virtualization is not enabled in the cloud. As Linux is required to run in virtualization as the default deployment for IT and the cloud, the virtualization feature such as KVM or Xen is required to be available and optimized in guests as well. . In this talk, we discuss use cases, characteristics of nested virtualization, and optimizations using new hardware virtualization features, such as “VMCS shadowing”. Also, we will provide some performance data and discuss the current performance bottle neck, the challenges it faces and look forward to how the nested virtualization to address those challenges.

Speakers

Bandan Das

Software Engineer, Red Hat

Bandan works on Virtualization at Red Hat. He is primarily interested in systems security and performance. Bandan has presented on various topics such as KVM, usb-mtp emulation in Qemu and the IIO interface in the Linux kernel.

Jan Kiszka

Principal Key Expert, Siemens

Jan Kiszka is working as consultant, open source evangelist and Principal Key Expert Engineer in the Linux Expert Center at Siemens Technology. He is supporting Siemens businesses with adapting, enhancing or strategically driving open source as platform for their product demands... Read More →

Yang Zhang

Intel

Yang is a software engineer from Intel. He has worked in virtualization field for five years.

Wednesday October 15, 2014 10:00am - 10:45am CEST
Room 3

KVM

10:45am CEST

Break

Wednesday October 15, 2014 10:45am - 11:15am CEST
Foyer

11:15am CEST

Towards Multi-Threaded Device Emulation in QEMU - Stefan Hajnoczi, Red Hat

Over the past year core parts of QEMU have been extended to support multi-threaded device emulation. This has enabled a rewrite of the experimental virtio-blk dataplane feature so that multi-threaded device emulation can become a fully-supported feature of QEMU. Although work is still ongoing, virtio-blk device emulation can now run in dedicated threads, called iothreads. These threads can be bound to host CPUs much in the same way that vCPUs can be bound. This makes it possible to achieve better scalability on multicore hosts where QEMU's legacy global mutex presents a bottleneck. This talk covers new thread-friendly infrastructure in QEMU and how it is used in virtio-blk. It also presents performance results showing the effect of multi-threaded device emulation. Finally, it outlines the remaining work and the new possibilities for multiqueue devices.

Speakers

Stefan Hajnoczi

IBM's Linux Technology Center

Stefan Hajnoczi has contributed to QEMU since 2010. He currently co-maintains the QEMU block layer together with Kevin Wolf. Now at Red Hat and previously at IBM's Linux Technology Center, Stefan has worked on disk image formats, storage migration, and I/O performance optimization... Read More →

Wednesday October 15, 2014 11:15am - 12:00pm CEST
Room 3

Storage

12:00pm CEST

Memory Externalization With userfaultfd - Andrea Arcangeli & Dr. David Alan Gilbert, Red Hat

Postcopy migration prototypes have been proposed before, but despite showing an excellent runtime, they depended on a special kernel device driver to manage the guest memory. To remove the shortcomings of the special device driver, we designed a core memory magagement feature in the Linux Kernel to provide memory externalization to qemu/KVM (and any other Linux application). The userfaultfd syscall combined with remap_anon_pages allows to efficiently handle page faults in userland, without modifying VMAs and without relaying on signal delivery. After this kernel feature was introduced, not a single line of KVM kernel code had to be modified in order to implement postcopy migration in qemu. This presentation will show the design of this work and statistics about the postcopy live migration runtime using userfaultfd. It will also cover some non-virt potential usage of the feature.

Speakers

Andrea Arcangeli

Red Hat

Andrea Arcangeli joined Red Hat in 2008 because of his interest in working on the KVM virtualization project, with a special interest in virtual machine memory management. Before Red Hat, he worked for SUSE for 9 years. He worked on many parts of the Linux Kernel, but especially on... Read More →

Dr. David Alan Gilbert

Biography coming soon.

Wednesday October 15, 2014 12:00pm - 12:45pm CEST
Room 3

Kernel

12:45pm CEST

Lunch

Wednesday October 15, 2014 12:45pm - 2:00pm CEST
Foyer

2:00pm CEST

Deterministic Replay and Reverse Debugging in QEMU - Pavel Dovgalyuk, ISP

We want to present our implementation of deterministic replay for qemu. Deterministic replay can be used for replay and reverse debugging of guest code through gdb remote interface. In this presentation we will review the challenges we faced while creating the deterministic replay, architecture of our solution of these problems, and applications of deterministic replay. Possible applications will include reverse debugging of the whole system including the kernel and user code.

Speakers

Pavel Dovgalyuk

Institute for System Programming

I am software deverloper in Institute for System Programming (ISP) of the Russian Academy of Sciences (RAS). The activities of the Institute include fundamental research, software development, applied research for the benefits of the Industry, and education. For the last 4 years I... Read More →

Wednesday October 15, 2014 2:00pm - 2:30pm CEST
Room 6

Migration

2:00pm CEST

Virtio-blk Multi-queue Conversion and QEMU Optimization - Ming Lei, Canonical

It is natural to map virtio-blk's virtqueue to blk-mq's hardware queue when blk-mq is ready in linux kernel now, then I/O request won't be blocked inside VM any more, but QEMU may be the next bottleneck of the whole I/O path. This presentation will introduce Ming Lei's in-progressing work on conversion to blk-mq of virtio-blk in both linux kernel and QEMU side, investigations why QEMU becomes new bottleneck, and proposed optimizations on QEMU to improve virtio-blk dataplane performance in case of multi virtqueue. These optimizations are based on model in which linux-aio may get best performance. Test result shows both throughput and scalability get lots of improvement about virtio-blk dataplane with both multi-queue conversion and QEMU optimizations.

Speakers

Ming Lei

Canonical

My name is Ming Lei, and I am a experienced linux kernel developer, have contributions to Block IO, USB stack, PM and other kernel components. Recently days, I am especially interested in Block I/O, kernel AIO and Virtualization, and have contributions to QEMU too. I am a software... Read More →

Wednesday October 15, 2014 2:00pm - 2:30pm CEST
Room 3

Storage

2:30pm CEST

Migration: Trying to Make it More Robust - Juan Quintela, Red Hat

Migration on kvm is known to be a fragile beast. Changing anything caused migration to break. Bigger problem was that we never detected that anything was broken until it was too late (i.e. after doing a release). This year has seen several tries to help improve this: - continuous vmstate testing: GSOC project, it allows us to check that we can load device state at any single moment - migration format checker (Amit): being able to see that the fields for each device help verify that migration would not fail before even attempted it - size array checks all over the tree (mst). - vmstate tests: We have tested every component on vmstate, This make much easier to do changes at the core, because we can check if the format has changed or not just with running a check. - more ....

Speakers

Juan Quintela

Red Hat

Born in Galicia (Spain). Starting doing PhD in Computing Science (Functional Programming). After 3 years, Linux Kernel Programming lured him. Worked at Mandriva as Kernel Maintainer from 2000 to 2005. In 2005 he joins the Virtualization Team at Red Hat. Currently he is working at... Read More →

Wednesday October 15, 2014 2:30pm - 3:00pm CEST
Room 6

Migration

2:30pm CEST

More Block Device Configuration - Kevin Wolf, Max Reitz, Red Hat

The traditional way of configuring block devices in QEMU leaves a lot to be desired. Last year's presentation outlined its problems, what a better mechanism might look like and the first steps towards it. After a year, a lot of progress has been made, and at the same time new challenges have become apparent. We will present the current state of affairs, take a look at the next steps to be done and possibly also yet unsolved design challenges.

Speakers

Max Reitz

Red Hat

Kevin Wolf

Red Hat

Kevin Wolf works at Red Hat as a KVM developer, with a focus on block devices. He is the maintainer of QEMU's block subsystem and has contributed many patches to block device emulation and image format drivers. After graduating in Software Engineering at the University of Stuttgart... Read More →

Wednesday October 15, 2014 2:30pm - 3:00pm CEST
Room 3

Storage

3:00pm CEST

Break

Wednesday October 15, 2014 3:00pm - 3:30pm CEST
Foyer

3:30pm CEST

Job Safety: Blockers in the Block Layer - Jeff Cody, Red Hat

QEMU supports running multiple block jobs, issued via QMP, while the guest continues to operate live. Not all operations can safely run concurrently, however. There is some support currently in QEMU to block unsafe operations, but the support is incomplete, and does not cover all scenarios. There are some proposed modifications to fix the gaps, and to increase the granularity of the block job locking. This talk will cover the existing job blocker support in QEMU, as well as new proposed methods to enhance the current support.

Speakers

Jeff Cody

Red Hat

Wednesday October 15, 2014 3:30pm - 4:00pm CEST
Room 3

Storage

3:30pm CEST

KVM on Grid, Shaken Not Stirred - Raphael Sack, Intel IT

Let’s say you have 10,000 Linux servers. Now, let’s say you need to upgrade these servers to version+1. How long is this effort going to take? Now consider the following scenario: All these servers are running 24/7 business critical tasks. Any disruption will cause significant delays in project timetables, and incur financial costs. How long would it take you to upgrade these servers now? You may decide the entire operation is unfeasible. KVM on grid is a flexible, fast solution enabling dynamic management and allocation of virtual machines in large scale high-performance environments. It revolves around state of the art open-source virtualization, combined with distributed computing models. This allows near-real time deployment of systems on a mass scale, minimizing the overhead and complexity of the environment. It’s not cloud, it’s not batch, it’s “KVM on grid”!

Speakers

Raphael Sack

Intel IT

Raphael Sack is a virtualization expert, system administrator and experienced programmer in the Engineering Computing group of Intel IT. Raphael has architected a number of Linux based solutions for high performance compute environments inside Intel. Additionally, he has written a... Read More →

Wednesday October 15, 2014 3:30pm - 4:00pm CEST
Room 6

User

4:00pm CEST

BoF / Hacking Time

Wednesday October 15, 2014 4:00pm - 7:00pm CEST
Room 3

BoF Hacking

9:00am CEST

Keynote: libvirt - Jiři Denemark

Speakers

Jiři Denemark

Red Hat

Jiri joined Red Hat to help maintain Xen in Red Hat Enterprise Linux 6 years ago. After one year he moved to the libvirt team at Red Hat where he works as a Senior Software Engineer and maintainer of libvirt packages. Jiri's main focus areas are guest CPU specification and mig... Read More →

Thursday October 16, 2014 9:00am - 9:15am CEST
Room 3

Keynote Management

9:15am CEST

Connecting High Performance End Users with Leading KVM Developers

As KVM is increasingly adopted by businesses and other organizations for virtualization, and is the most popular hypervisor for OpenStack deployments, there is real value in bringing together high performance end users with the leading KVM community developers.

This panel of KVM end users aims to encourage these conversations by sharing experience of KVM practical deployments, KVM in new technologies, identifying KVM technical needs, and discussing how developers and end users can work together to help make KVM better.

Moderated by the Open Virtualization Alliance, this session will also explore how best to enable cutting-edge end users to network together and with the development community.

Thursday October 16, 2014 9:15am - 10:00am CEST
Room 3

User

10:00am CEST

OpenStack Performance Optimization with NUMA, Huge Pages and CPU Pinning - Daniel Berrange, Red Hat

The OpenStack project builds and distributes the leading open source cloud infrastructure platform. Of all the hypervisors supported, KVM is by far the most widely deployed in production, by an order of magnitude. In the most recent dev cycles there has been a big uptake in interest in optimizing hardware performance. To that end there is work being done to enable full advantage to be had from NUMA hardware, huge pages and CPU affinity. This talk will look at the work undertaken and the unique challenges faced in a cloud environment, as opposed to traditional data center virtualization.

Speakers

Daniel Berrange

Red Hat

Daniel is a Principal Software Engineer, working in a variety of roles at Red Hat over the last 12 years. Since 2006, he has been specialized in the development of technologies related to virtualization management, as lead developer of Libvirt, GTK-VNC, Libvirt Perl, Libvirt GObject... Read More →

Thursday October 16, 2014 10:00am - 10:45am CEST
Room 3

Management

10:45am CEST

Break

Thursday October 16, 2014 10:45am - 11:15am CEST
Foyer

11:15am CEST

Thanks for Live Snapshots, Where's Live Merge? - Adam Litke, Red Hat

oVirt has supported the ability to take live snapshots of a VM since version 3.2 but these snapshots could only be deleted or merged when the VM is powered off. On the surface this seems like a small incongruity that should be easily fixed but it has taken us three major releases to deliver live merge capabilities. This talk will demonstrate how each component in the KVM stack works together to seemlessly merge a snapshot and how oVirt ensures reliability and data integrity when facing storage connectivity issues and other unexpected events. Next, we will cover how live merge improves other oVirt features such as VM Backup and Live Storage Migration. Finally, a discussion of new ideas and future improvements will be presented.

Speakers

Adam Litke

Red Hat

Adam has been involved in the Linux community since 1999 where he worked with an IBM team to boot Linux on the ppc64 architecture for the first time. Since then he has contributed to many projects including: kernel, libhugetlbfs, Memory Overcommitment Manager, qemu, libvirt, and oVirt... Read More →

Thursday October 16, 2014 11:15am - 12:00pm CEST
Room 3

Management

12:00pm CEST

Libvirt. Why Should I Care? - Michal Privoznik, Red Hat

QEMU is a beautiful part of the KVM virtualization stack. It runs quickly, scales beautifully and is portable. However, it has some limitations that libvirt covers up. Things that libvirt does on top of QEMU are shown. The talk will step the audience through the process of starting a new VM, where libvirt work is shown the best. Other areas like networking, cgroups, secrets and snapshots are covered, too. The aim is to show what an important role libvirt plays in the virtualization stack.

Speakers

Michal Privoznik

Red Hat

I am one of main developers of the Libvirt project. and currently working for Red Hat for 3 years where I started during my PhD studies. Last year I gave talk on LinuxCon in Tokyo, Japan. Beside my regular work on libvirt, I maintain libvirt-snmp, contribute to other projects, and... Read More →

Thursday October 16, 2014 12:00pm - 12:45pm CEST
Room 3

Management

12:45pm CEST

Lunch

Thursday October 16, 2014 12:45pm - 2:00pm CEST
Foyer

2:00pm CEST

I/O Demand-driven VM Scheduler in KVM - Hyotaek Shim, Samsung Electronics

Although virtualization technology allows a high degree of server consolidation to maximize the resource utilization, it results in performance interference between consolidated servers running on the same physical machine, where CPU-intensive VMs and I/O-intensive VMs often coexist. We have observed not only that computing-intensive VMs still interfere with co-running I/O-intensive VMs, but also that I/O performance rapidly drops when each VM performs a mixed workload of computing and I/O. To handle this problem, we develop an I/O demand-driven VM scheduler, which enables KVM to track the characteristics of VM’s tasks and to preempt computing tasks in a VM for timely boosting I/O-intensive tasks in other VMs. Our experimental results show that the proposed scheduler highly improves the I/O performance of VMs, but causes negligible degradation in the computing performance.

Speakers

Hyotaek Shim

Samsung Electronics

Hyotaek Shim received his M.S.-Ph.D. joint degree in Computer Science from Korea Advanced Institute of Science and Technology (KAIST) in 2013. His research areas include virtualization systems, cloud computing systems, and storage systems. He is currently a senior research engineer... Read More →

Thursday October 16, 2014 2:00pm - 2:30pm CEST
Room 3

Kernel

2:30pm CEST

KVM vs. Valgrind - Christian Bornträger, IBM Germany

valgrind is an widely used award-winning debugging tool. This talk will focus on using valgrind with KVM/QEMU. To understand what works and what not, a basic overview on how valgrind works internally is given. Furthermore real examples of qemu bugs, fixes for valgrind and usage hints are given to illustrate how to use valgrind in a virtualization environment.

Speakers

Christian Bornträger

IBM Germany

Christian Bornträger is a software engineer at IBM germany and is the s390 maintainer for KVM and the s390/KVM related parts of QEMU.

Thursday October 16, 2014 2:30pm - 3:00pm CEST
Room 3

QEMU

3:00pm CEST

Break

Thursday October 16, 2014 3:00pm - 3:30pm CEST
Foyer

3:30pm CEST

QOM Exegesis and Apocalypse - Paolo Bonzini, Red Hat

The QEMU object model was first introduced into QEMU in 2011. The way it's been adopted since then, however, doesn't always match the originally planned shape of QOM. This talk will show how QOM integrates with other infrastructural components of QEMU (most notably QAPI), where and how QOM ended up being used, and propose ideas for its future evolution.

Speakers

Paolo Bonzini

Red Hat

Thursday October 16, 2014 3:30pm - 4:00pm CEST
Room 3

QEMU

4:00pm CEST

Testing QEMU Emulated Devices Using qtest - Marc Marí Barceló, UPC BarcelonaTech

QEMU as any other software project needs tools for continuous checking and testing to avoid inserting bugs in already working code. This need is increased by its active development and frequent releases. Due to its high number and heterogeneity, emulated devices and supported drivers are especially important to test and check every new build. To simplify the development of this tests to the device developers, a simple but really powerful testing suite framework, called qtest, has been added to QEMU in the last years. In this presentation, Marc Marí will show how to develop tests for devices in QEMU, what drivers are available at the moment, how to debug the tests and how to share it with the community.

Speakers

Marc Marí Barceló

UPC - BarcelonaTech

Student of Informatics Engineering specialized on Computer Engineering at Technical University of Catalonia (UPC - BarcelonaTech). GSoC student during summer 2014 in QEMU project working on Device driver frameworks. Software developer at UPC projects such as 3Cat-1 nano-satellite... Read More →

Thursday October 16, 2014 4:00pm - 4:30pm CEST
Room 3

QEMU

4:30pm CEST

Group Picture

Thursday October 16, 2014 4:30pm - 5:00pm CEST
Room 3

5:00pm CEST

Closing Session

Thursday October 16, 2014 5:00pm - 5:30pm CEST
Room 3

6:00pm CEST

Hacking Time

Thursday October 16, 2014 6:00pm - 7:00pm CEST
Room 3

Hacking