AMD hosted the Xen Summit North America 2010 on April 28-29 in our Sunnyvale headquarters.

Welcome & Keynote

  • Episode 1: Xen Community Update (Ian Pratt) (53:33)
  • Episode 2: Xen Hypervisor Project Update (Keir Fraser) (20:53)
  • Episode 3: AMD, Xen, and Virtualization (Tom Woller) (25:27)

 


Energy Savings

  • Episode 4: SleepServer: A Software-Only Approach for Reducing the Energy Consumption of PCs within Enterprise Environments (Yuvraj Agarwal) (31:35)
    In this paper, we describe the architecture and implementation of SleepServer, a system that enables hosts to transition to low-power sleep states while still maintaining their application’s expected network presence using an on demand proxy server. Our approach is particularly informed by our focus on practical deployment and thus SleepServer is designed to be compatible with existing networking infrastructure, host hardware, operating system and application software and introduces only a trivial software agent on each system under management.
  • Episode 5: Energy-Efficient Storage in Virtual Machine Environments (Lei Ye) (22:48)
    This paper explores the disk I/O activities between VMM and VMs using trace driven simulation to understand the I/O behavior of the VM system. Subsequently, this paper proposes three mechanisms to address the isolation between VMM and VMs, and increase the burstiness of hard disk accesses to increase energy efficiency of a hard disk.


Xen Cloud Platform

  • Episode 6: Xen Cloud Platform Update (Jonathan Ludlum) (31:07)
    Update on what Citrix is doing and future directions.
  • Episode 7: Case Study: Iaas using XCP and XAPI (Marco Sinhoreli) (25:14)
    Globo.com is the Internet branch of Organizações Globo, the biggest media conglomerate in Latin America, with offices in Rio de Janeiro and São Paulo. It aggregates the greatest web vertical portals in Brazil, spreading from News (G1), Sports (globoesporte.com), Videos (Globo Videos) to Celebrities (Ego). Globo.com IaaS chose the Xen Cloud Platform for its paravirtualization, complete API for storage, pooling, network management and community support features.
  • Episode 8: VastSky – Cluster Storage System for XCP (Hirokazu Takahasi) (31:42)
    VastSky is a cluster storage system comprised of commodity hardware — PC servers and SATA disks. The systems are designed to be used in a cloud environment, which is scalable and fault-tolerant, and has a feature that virtual machines can directly run on the system. This presentation discusses the concept, design and roadmap of VastSky.
  • Episode 9: Building an Infrastructure as a Service Cloud on XCP (Sheng Liang) (23:25)
    This talk discusses how an Infrastructure as a Service (IaaS) cloud can be built on the Xen Cloud Platform (XCP), including how to scale servers by grouping them into availability zones and pods, organize multiple primary VM disk storage servers and secondary storage servers, and manage isolated guest networks using layer-2 tunneling and hardware VLANs. It also covers service management features such as service definition, usage metering, and OSS/BSS integration.
  • Episode 10: XRM: Event-based Resource Management Framework for XCP (Pradeep Padala) (29:39)
    Xen Cloud Platform (XCP) is emerging as an open source solution to build Infrastructure as a Service (IaaS) platforms. XCP is still in its infancy, and one of the important missing features is “resource management” in virtualized data centers. In this work, we propose XRM, a resource management (RM) framework for XCP that provides a modular and extensible framework to implement RM strategies for load balancing, optimal VM placement for high utilization, optimal power utilization, and high application performance.


Project Updates

  • Episode 11: PVOps Update (Konrad Rzeszutek Wilk) (15:32)
    Differences between XenLinux and PVOps, and the advantages of PVOps.
  • Episode 12: Xen Scheduler (George Dunlap) (22:18)
    Overview of project, design target, and development plans.
  • Episode 13: PCI Express Support in QEMU (Isaku Yamahata) (26:18)
    Currently in QEMU, passing through of PCI is supported and PCI express devices can be passed through as PCI devices but not as PCI express natively. PCI express has more features than PCI, such as MMCONFIG, native hot plug (not ACPI based), ARI(Alternative Route ID), AER(Advanced Error Reporting) and more. Development issues for PCI express are discussed, including needed enhancements in QEMU and BIOS.
  • Episode 14: Xen ARM Update (Sang-bum Suh) (20:36)
    Current hardware and software support, latency issues to be improved, real-time requirements.


Clouds

  • Episode 15 (two parts): GoGrid and Xen (Paul Lappas) (12:06 / 24:21)
    How GoGrid uses the Xen hypervisor, trends that we see in the market around hypervisor adoption, and the opportunities they provide.
  • Episode 16: Open Source Cloud Computing (Bernard Golden) (29:48)
    Why open source is crucial to the future of cloud computing.


Real-Time

  • Episode 17: Supporting Soft-Real-Time Tasks in the Xen Hypervisor (Shalini Yajnik) (42:11)
    Soft real-time applications, such as media-based ones, are impeded by components of virtualization including low-performance virtualization I/O, increased scheduling latency, and shared-cache contention. The virtual machine scheduler is central to all these issues. The goal in this paper is to adapt the virtual machine scheduler to be more soft-real-time friendly by improving two aspects of the VMM scheduler – managing scheduling latency as a first-class resource and managing shared caches.
  • Episode 18: Extending Xen into Embedded and Communications Applications (Edwin Verplanke/Don Banks) (29:11)
    Today’s VMM’s and silicon architecture generally cater very well to general purpose environments and applications; however, the embedded and communications environments require enhanced functionality such as real-time scheduling, high performance I/O, high availability, and co- located cooperating applications. This presentation will cover some new and exciting virtualization usage models – does Xen have what it takes to address embedded and communications requirements?


Miscellaneous Topics

  • Episode 19: Graphics Passthrough Challenges (Allen Kay) (52:07)
    A description of the fundamentals of Xen HVM PCI passthrough as it works today in upstream Xen for devices such as NIC and USB controllers, plus details on special challenges and enhancements necessary for bringing up discrete graphics controllers and various generations of Intel integrated graphics devices in the guest environment.
  • Episode 20: Xen NUMA Guests (Dullor Rao & Jun Nakajima) (33:07)<
    Power and performance constraints are pushing platforms towards increasingly NUMA architecture. While such platform architectures provide greater aggregate bandwidth and are more scalable, they also necessitate changes in the system software for optimal performance. But when running on top of a VMM, the domains are completely unaware of the underlying asymmetry, leading to unpredictable performance overheads. In this presentation, we discuss the cost of virtualization on such platforms. We also present a global domain memory allocation scheme for Xen, the implementation of the allocation scheme with PV NUMA guests, and more.
  • Episode 21: Application of Fuzzy Control Theory to Resource Management in a Virtualized System (Sho Niboshi) (52:50)
    Nowadays, virtualization technology is widely used to reorganize data centers and server systems with a small number of computers by incorporating multiple systems into a single physical computer. On the shared system, the resource controller controls resource assignment to virtual machines (VMs) and plays an important role in determining the virtualized system’s performance. However, the resource controller in current systems does not have any guarantees for application performance because the allocation function only utilizes the information from the VM instead of the applications themselves. This paper demonstrates a resource controller that takes application state, e.g. Quality of Service (QoS), into account to identify resource demand. We applied fuzzy control theory for the resource allocation to model the complex relationship between QoS and demand. We evaluate the fuzzy rule-based controller  on a Xen-based system with two guest VMs running mail and Java application servers, and show its advantages over the Xen default scheduler.
  • Episode 22: Update on Transcendent Memory in Xen (Dan Magenheimer) (20:44)
    At Xen Summit 2008, we described self-ballooning. At Xen Summit 2009, we introduced Transcendent Memory (TMEM). With Xen 4.0, we have combined the two into a unique enterprise-ready memory utilization optimization solution. For Xen Summit 2010, we review the goals and basics of the two topics, discuss significant advances in tmem both in Xen and in Linux, and present some new performance results.
  • Episode 23: Neon: System Support for Derived Data Management (Qing Zhang) (21:42)
    Modern organizations face increasingly complex information management requirements. Among these, personally identifying customer records must be carefully access-controlled, sensitive files must be encrypted on mobile computers to guard against physical theft, and intellectual property must be protected from both exposure and “poisoning.” However, enforcing such policies can be quite difficult in practice since users routinely share data over networks and derive new files from these inputs—incidentally laundering any policy restrictions. In this paper, we describe a virtual machine monitor system called Neon that transparently labels derived data using bytelevel “tints” and tracks these labels end to end across commodity applications, operating systems and networks. Our goal with Neon is to explore the viability and utility of transparent information flow tracking within conventional networked systems when used in the manner in which they were intended. We demonstrate that this mechanism allows the enforcement of a variety of data management policies, including data-dependent confinement, mandatory I/O encryption, and intellectual property management.
  • Episode 24: Xenalyze: Analyze Xen Traces (George Dunlap) (18:52)
    Xen’s trace infrastructure can produce a wealth of information about the execution of a running Xen system, useful for profiling, debugging, or just figuring out what’s going on. However, sorting through that data and making sense of it is a much more difficult matter. Xenalyze is a tool I’ve been developing over the last three years to make sense out of the data. Its first big feature is its attempt to reconstruct the order that traces occurred originally across multiple processors, even in the face of clock skew and lost records. The second is to track individual vcpus as they migrate across physical cpus, collecting statistical information about them. Finally, it can collect statistical information and display it in a “summary” form (across the whole run), various graphs, or a record-by-record exposition. The talk will briefly describe Xen tracing infrastructure, the xenalyze tool, and various uses to which it can be put.
  • Episode 25: Evolving New Configuration Tools for IOV Network Devices (Mitch Williams) (34:42)
    I/O Virtualization (IOV) technology for network devices is still in its infancy. While the devices are readily available, and drivers have been pushed into the kernel, configuration tools are few and far between. Kernel maintainers and network administrators are still coming to terms with what types of tools are required to make IOV network devices usable in the real world. This paper describes the current state of these configuration tools, shows some use cases, and provides an overview of future development work. It describes what works today, what’s still missing, and what can be done to address these issues.
  • Episode 26: Libxenlight (Stefano Stabellini) (27:19)
    Many different tool stacks are currently used to manage a Xen-based host, leading to inconsistencies, code duplications and bugs. Moreover Xend, the default tool stack on xen-unstable, is hard to modify and extend. Libxenlight was created to fix these issues. This talk explains the design principles, the architecture and the objectives of this new library. Libxenlight aims to provide a simple and robust API for tool stacks to do Xen operations and to create a common codebase for the lower-level implementation of all the various Xen tool stacks.
  • Episode 27: Closing Presentation (Ian Pratt) (26:12)