This event has ended. Visit the official site or create your own event on Sched.
October 23-26, 2017 - Prague, Czech Republic
Click Here For Information & Registration
View analytic

Sign up or log in to bookmark your favorites and sync them to your phone or calendar.

Joint KVM Forum Track [clear filter]
Wednesday, October 25


Lessons in Running libvirtd at Scale - Prerna Saxena, Nutanix
At Nutanix, we design for scale, and the efficiency of the libvirt/QEMU stack is key to our overall throughput. We primarily consume libvirt qemu driver via a proprietary, distributed orchestrator. We target hyperscale of 1000+ VMs per host. We found that the current handling of QMP asynchronous events has caused the daemon to sporadically lock up under stress. This talk covers ongoing community efforts[1] to improve this. We also advocate scalability improvements with better RPC queuing and streamlined requests. The current queuing model in libvirtd can be overwhelmed with patterns of RPC requests, compromising throughput. Asynchronicity of the daemon is flipped in the current libvirt-client implementation, causing RPCs to appear blocking which they aren't. This talk explores these aspects in detail. [1] https://www.redhat.com/archives/libvir-list/2017-May/msg00016.html

avatar for Prerna Saxena

Prerna Saxena

Prerna Saxena implemented libvirt support for PowerPC architecture, and has also been associated with QEMU tracing implementation in the past.In her current role at Nutanix, she works for the Acropolis Virtualization platform, ensuring accurate and efficient management of VMs wit... Read More →

Wednesday October 25, 2017 11:15 - 11:55
Grand Ballroom


OMG, NPIV! Virtualizing Fibre Channel with Linux and KVM - Paolo Bonzini, Red Hat, Inc. & Hannes Reinecke, SUSE
Fibre Channel is a widely deployed SAN technology. It does provide for a native virtualization support in the form of NPIV (N_Port Id Virtualisation). Unfortunately the Linux implementation, which presents the NPIV port as a SCSI Host, does not lend itself easily to virtualisation with QEMU. Currently QEMU can only forward individual PCI devices and block devices, but no easy way exists to specify how NPIV devices can or should be forwarded. Furthermore, all devices and possibly the partitions and filesystems therein are exposed to the host, which may cause security concerns. This talk will describe the possible solutions for FC virtualization at both the Linux and QEMU levels, and their relative advantages and disadvantages. A short introduction to FC concepts will be included, covering the protocol, HBA and operating system levels.

avatar for Paolo Bonzini

Paolo Bonzini

Principal Software Engineer, Red Hat, Inc.
I have 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 first presented my work on QEMU... Read More →
avatar for Hannes Reinecke

Hannes Reinecke

Teamlead Storage & Networking, SUSE Linux GmbH
Studied Physics with main focus image processing in Heidelberg from 1990 until 1997, followed by a PhD in Edinburgh 's Heriot-Watt | University in 2000. | Working at SUSE Labs with focus on storage and mainframe. Principal contact point for storage related issues on SLES and te... Read More →

Wednesday October 25, 2017 12:05 - 12:45
Grand Ballroom


QEMU in UEFI - Alexander Graf, SUSE
UEFI is a firmware specification created by Intel with portability in mind. The UEFI way of doing that was to provide special UEFI byte code (EBC). Unfortunately nobody really cared so compiler, firmware support and providers of EBC option roms ceased and basically every UEFI option rom today contains native x86(_64) machine code. If you now want to plug a PCIe card into your shiny ARM server, that means even though firmware would be compatible it still can't execute the option rom. Until you add QEMU to the mix. Join me in exploring the depth of UEFI binary interfaces, marshalling between different architecture's function call ABIs on the fly and learn how to integrate all of this into a working firmware, running on real hardware, driving a real PCIe adapter.

avatar for Alexander Graf

Alexander Graf

KVM Wizard, SUSE
Alexander started working for SUSE about 9 years ago. Since then he worked on fancy things like SUSE Studio, QEMU, KVM and openSUSE on ARM. Whenever something really useful comes to his mind, he tends to implement it. Among others he did Mac OS X virtualization using KVM, nested... Read More →

Wednesday October 25, 2017 14:15 - 14:55
Grand Ballroom


The Future of virtio: Riddles, Myths and Surprises - Jens Freimann, Red Hat, Inc.
As many guests use virtio for all their IO needs, the performance of virtio puts a hard limit on the system performance. virtio is useful as it's a hardware-independent interface - but for some guests, that independence comes at a performance cost. That's why for the last year the virtio community has been looking at different ways to extend virtio - making it work on the underlying hardware better, but without breaking the indepence. This work made us re-examine several underlying assumptions made during early stages of the design of the virtio ring. Some of the findings from this re-examination were surprising; some of the common assumptions are a myth. This presentation is an update on the progress made on the next version of virtio and its future - it will try to dispel some myths and describe some things about virtio that puzzled us and some things we found out that surprised us.


Wednesday October 25, 2017 15:05 - 15:45
Grand Ballroom


Helping Users Maximize VM Performance - Martin Polednik, Red Hat, Inc.
QEMU supports numerous options to fine tune the virtual machine, starting from the big items such as number of CPUs, NUMA nodes or amount of memory all the way down to the choice of USB controllers, disk controllers, and thread pinning. Different use cases require careful tuning to reach desired performance, and certain settings may interfere with configuration's expected performance. Management software can let users fully build the virtual machine, but it may also help them by warning about conflicting or suboptimal choices. In this talk, we will present samples of common configurations, some of them from oVirt users' community, and our ideas how to make their life easier by suggesting changes that should lead to performance improvement in most cases.

avatar for Martin Polednik

Martin Polednik

Software Engineer, Red Hat
Martin Polednik works on the oVirt project as a Software Engineer at Red Hat. As part of the oVirt virtualization team, he is responsible for integrating KVM, QEMU and libvirt virtualization features into oVirt.

Wednesday October 25, 2017 16:15 - 16:55
Grand Ballroom


KVM Performance Tuning on Alibaba Cloud - Yang Zhang, Alibaba Cloud
Millions of users are deploying their services in Alibaba Cloud which based on KVM. The performance and capacity of KVM are two critical indicator for the whole system. In this presentation, Yang will demonstrate some real performance issues that reported by end users pertained to KVM, and the structural way to analyze and solve the problems. The performance tuning involves timer, IPI, memory and scheduler. In this implementation, lots of KVM specified features like exit-less timer, PV interrupt and VCPU ware scheduler in Linux kernel and KVM are systematically developed and deployed, which will be submitted to KVM community soon. At the end, numbers of performance improvements data of Alibaba Cloud after the overall optimization are showed in the presentation, and certain open-loop issues that still not handled well in KVM which are important to end customers are raised up to audiences.


Yang Zhang

Yang is an expert in virtualization field. He first participated in Xen and KVM community in 2008. Most of his contribution was related on Xen and KVM part. He had been a maintainer on Xen VT-D component for two years and the main contributor on Xen nested virtualization and KVM... Read More →

Wednesday October 25, 2017 17:05 - 17:45
Grand Ballroom