KVM has brought Linux to the forefront of data center virtualization and cloud computing. The KVM project evolves rapidly to meet the challenges of these computing environments. Come learn about the advances KVM has made over the past year, and what new developments are on the horizon for KVM. The session will discuss the major new developments in KVM/QEMU in the last year, and in particular those targeting data centers and cloud management.The audience should be familiar with server virtualization technology, general data center and cloud computing concepts.
Recently Software Defined Network (SDN) technology has become important and openflow is focused on as basic building block of SDN. Those technologies are important for creating test bed for experimental protocol by standardizing protocol to manage switches and production use to reduce maintenance cost of network by automation via software as network virtualization. As they are emerging technology, the development of software to use openflow is hot area. In this presentation, I review SDN, openflow and openvswitch which is opensource software implementation of openflow. And then survey openflow controllers describing their features and goal. The target audience is those who would like to take overview of those technology. No deep knowledge is required.
Storage management is one of the hardest parts of running a virtualized data center. Storage is slow but large; storage holds precious data and at the same time it is subject to disastrous failures. Due to low speed and large size, it is often impossible to keep virtual machines offline while doing maintenance on the host's disks. Yet, virtualization promises to minimize your downtime and even to help you delegate maintenance tasks. In order to reconcile these two opposites, live disk operations let you perform maintenance while the virtual machine is running. Support for these operations is a recent addition to the QEMU/KVM storage subsystem; this talk will cover the various tasks that QEMU can perform live, how it does them, and why you shouldn't be afraid of it.
How to Mitigate Latency Problems during KVM/QEMU Live Migration
This presentation first introduces the basics of KVM/QEMU live migration focusing on how to synchronize guest memory between the running VM and target VM on a destination node, and explains latency problems thereof with actual data which clearly show how long the guest's VCPUs could be forced to wait for a lock. It is then followed by the latest improvements and developments for the future. The expected audience includes both Linux developers and users, especially who are interested in virtualization technologies: you can concentrate on some topics you are interested in without fully understanding the details you are not familiar with. Based on his experience, including both technical discussions and communication with end users, the speaker will do his best to make this talk to lead to fruitful discussions that would benefit further development of KVM/QEMU.
Yabusame: Postcopy Live Migration for QEMU/KVM
Postcopy live migration is yet another migration mechanism that allows users to change the execution host of a VM within one second while keeping visible disruption to a minimum. In addition, the whole migration process is basically shorter than normal live migration. It will provide great benefits for load balancing and energy savings using VMs. Although an implementation for academic research was available, it wasn't designed for the upstream merge and production use. So I redesigned/implemented it. Especially it takes advantage of asynchronous page fault features which can't be utilized by precopy approach. In this talk, I explain the new design and implemetation for QEMU/KVM live migration which is production-ready and show the evaluation results with demo movie. The target audience is virtualization developers and advanced users who is looking for new features.
OpenFlow, software defined networking, and network virtualization have become hot topics for networking and cloud computing. Open vSwitch is designed to address challenges in these areas by enabling the network to be more programmable than it has been traditionally. Programmatic control is a necessity in virtualized environments as endpoint locations become dynamic and large scale operators expect an API to their data centers. The features and architecture of Open vSwitch will be covered as well as how they can be applied to the problem of software defined networking. This presentation is aimed at both networking developers and deployers. For developers it will provide an overview of how Open vSwitch can be programmed to automate networks, while end users will learn about the available capabilities and what to expect when using them.
Status and Directions For Libvirt Virtualization API
The topic of this talk is to present the current status of libvirt, its relationship with others virtualization projects like Ovirt and OpenStack. Then a not too deeply technical discussion about the direction of the current project evolution, among the topics would be:
Libvirt Towards Enterprise Use
libvirt is a toolkit to interact with the virtualization capabilities of Linux. It provides a set of C API by which one can interact with the underlying virtualization infrastructure easily. For enterprise use, there are a lot of libvirt-based system adminstration tools. Fujitsu made lots of efforts to make libvirt more flexible and powerful in the past years. Fujitsu also has some plans at libvirt on enterprise point of view.This talk will focus on enterprise customised features and improvements. For instants, accounting and resources control, enhancements of dump, enhancements of manuals, etc. The audience should be libvirt developers, system administrators or users interested in libvirt.
Low-Overhead Ring-Buffer of Kernel Tracing in a Virtualization System
"IVRing", a low-overhead ring-buffer to send tracing data from guest OSs to a host OS, is introduced in a virtualization system. IVRing does not use network I/O which causes a high overhead. IVRing takes advantage of IVShmem, which is a virtual RAM device, as a communication path for tracing data. Since IVShmem allows to sharing a memory region between guests and host applications via POSIX shmem, data passing is possible without memory copying. A writer on a guest OS records trace data to the ring-buffer. On the other hand, a reader on a host OS directly refers to a part of the ring-buffer and outputs tracing data. Therefore, there is no need to copy tracing data from guests to a host by using IVRing. This method helps for system monitoring such as in virtualized mission-critical systems or cloud systems.
Tracing Across Host OS and Guest OS
We introduce "Integrated tracing" that enables to trace a host OS and guest OSs transparently. This technology based on IVRing that is a ring-buffer allows copy-less data communication between a host OS and guest OSs. In the virtualized environment, it was too difficult to find the cause of failures. We had to analyze huge trace-data of all guest OSs and a host OS, because a guest behavior can be affected by other guests. In this case, it is difficult to find the relevance to the failure and its causes. Moreover, since each guest OS has unique time-stamp, it is impossible to merge without time-stamp correction. This presentation shows that how the integrated tracing correct time-stamps and visualize tracing data to help to clarify the relevance to the failure and its causes about I/O.
As KVM gains popularity in the cloud computing world, the ability to quickly locate and resolve performance issues in KVM becomes even more pronounced. In this presentation, I will present the tools, e.g. kvm_stat and kvm-events, that help quickly identify and resolve performance issues. I will also present the work I did to develop the tool. My presentation will be helpful to system administrator looking to tune their environment for optimal performance as well as developers interested in the internal working of this infrastructure.
The oVirt Project is an open virtualization project providing a feature-rich server and desktop virtualization management platform with advanced capabilities for hosts and guests, including high availability, live migration, storage management, system scheduler, and more. oVirt provides an integration point for several open source virtualization technologies, including kvm, libvirt, spice and oVirt node. oVirt was launched in November 2011 as a fully open source project, based on assets from Red Hat Enterprise Virtualization Manager platform. The project has an open governance model, and initial board has members from IBM, Cisco, Netapp, Red Hat and SUSE. The session will provide an intro to the project components and features
