Microvisor: A Scalable Hypervisor Architecture for Microservers Xenia Ragiadakou, Michail Alvanos, Julian Chesterfield, John Thomson, Michail Flouris OnApp Ltd, UK {firstname.lastname}@onapp.com

1

Introduction

Virtualization of server hardware is a commonly used practice to provide scalable resource management. There are many benefits to virtualizing hardware resources, primarily the ability to efficiently share resources (CPU, memory, NICs) across a multi-tenant platform hosting many Virtual Machines (VMs). Each VM presents a closed environment for booting a standard OS kernel and running applications within that OS container. In addition to commonly deployed commercial hypervisors (HVs), there are two dominant open-source virtualization platforms: Kernel-based Virtual Machine (KVM) [1] and the Xen Hypervisor [2]. The platforms have mature support for established Intel x86 architecture chipsets and are both making fast advances towards mature support for ARM architecture hardware. Traditional x86 server architectures, having reached frequency scaling limits, have improved performance through increasing the number of cache-coherent logical cores and available caches. To improve scalability in a powerefficient manner, recent server architectures implement non-uniform memory access (NUMA), where each CPU core is directly attached to a portion of memory, but is also able to address all system memory, albeit at different speeds. Due to the power and performance limitations, alternative architectures have emerged with non-cachecoherent memories between groups of processor cores [3] commonly referred to as Microservers. New OS system architectures are required to enable applications to execute across multiple, low-power, independent, non-cachecoherent islands (denoted as chiplets) with access to shared hardware I/O device resources. These architectures have better scaling properties that allow increasing numbers of CPU cores, while maintaining a high performance-topower ratio, which is the key metric if hardware is to continue to scale to meet the expected demand of ExaScale computing and Cloud growth. In contrast to the KVM Hypervisor platform, the Xen Hypervisor provides a true Type I Hypervisor. That is to say that the Hypervisor layer runs directly on the baremetal hardware, managing guest OS instances directly above it. There is no Host Operating system required and in this

way the Type I architecture is considered to be a minimal, high performance shim. In parallel to the virtualized guest systems running on the Type I Hypervisor, traditional Xen systems utilize a control domain, known as Dom0, which has privileged access to the hypervisor and is responsible for various tasks including; administering guest virtual machines, managing resource allocation for guests, providing drivers for directly attached hardware, and offering network communication support. Guest Virtual Machines (DomUs) do not typically have direct access to real hardware, and thus all guest domain network and storage communication is managed through paravirtual device drivers hosted in Dom0 which in turn handles safe resource access multiplexing of the physical hardware. In order to meet a variety of emerging technology trends, a novel, super-lightweight Type I Hypervisor architecture is essential. The proposed ‘Microvisor’ architecture is significant for a number of reasons. 1. Resource utilization efficiency. Microservers are servers with limited resources and high performanceto-power ratios, which are built and managed as a cluster of cache-coherent CPU islands. 2. Performance and scalability. To meet the performance and scale demands of emerging Network Function Virtualization (NFV) platforms, a significantly rearchitected paravirtual network IO path is required. Network performance between virtualized functions executing on commodity hardware platforms today is limited primarily due to context switches between the VM and the control domain that handles the packet switching. In order, therefore to meet the growing demand of Network Function Virtualization on commodity hardware, using individual worker VMs to provide enhanced network packet filtering, middlebox functionality and packet forwarding, requires significant changes to the architecture of commodity virtualization platforms. It is thus key for performance that the packet forwarding layer is as fast as possible, with slow path network functions offloaded to separate processing engines (SDN Network Function Virtualization model). We propose a new hypervisor design called Microvisor, which is under development. The platform is based on the

Open Source Xen Hypervisor but presents significant architectural changes. The core idea behind the Microvisor is to remove any dependency on a local Dom0 domain for inter-domain paravirtual device communication but support unmodified virtualized guest OS and kernels. Additional features include the support of VM setup, bootstrap and resource allocation without any control domain dependency. We move this generic functionality into the Xen Hypervisor layer directly whilst maintaining Hardware driver support through a super-lightweight driver domain interface. This results in an extremely efficient, minimal virtualization shim, that provides all the generic inter-domain virtualization functions.

1.1

Current Xen Architectural Design

The Control domain (Dom0) typically boots immediately after the Xen Hypervisor image. It provides hardware driver support for key shared IO resources (network and storage) and management tools to control VM lifecycle operations (start/stop/suspend/resume) for a guest domain as well as the ability to migrate the live state of a running guest to another physical hypervisor with only minor service interruption. The Control domain is also responsible for managing the Xenbus communication channel which is primarily used to signal virtual device information to a guest domain. It also manages the physical memory pool which is shared across VMs, and the virtual CPU to VM mapping by communicating a CPU scheduling algorithm to the HV. For every active paravirtualized device of a guest VM, there is a corresponding driver in the control domain that allocates resources and handles the communication ring over which virtual I/O requests are transmitted. The control domain is then responsible for mapping those I/O requests onto the underlying physical hardware devices, such as Ethernet frames over a NIC, or block I/O requests to an iSCSI block device. The drivers, known as frontend (guest-side) and backend (Dom0-side), are implemented using ring buffers between Dom0 and the associated DomU PV device instance.

1.2

Figure 1. The new Hypervisor Architecture. Remote management of a thin Microvisor is provided over a network interface. The long-term vision is that Microvisors will form a cluster of cooperative hypervisor shims that execute directly on baremetal hardware and can be remotely managed as simple commodity raw Ethernet appliances, or via custom hardware interfaces for more complex Microserver hardware. In the former case, secure access is physically enforced using Layer 2 Ethernet mechanisms to ensure that only trusted elements can communicate directly with the Microvisor, as shown in Figure 1.1. A simple packet forwarding soft switch is embedded directly in the Xen Hypervisor layer which provides simple and fast packet forwarding functionality between frontend guest netfront instances using a zero-copy mechanism. Packet forwarding off the host is provided via a hardware driver domain, plugged directly into the logical soft switch. VM Block storage is provided either by access to persistent direct attached storage drives, or via network-attached storage protocols (e.g. iSCSI, ATAoE). In both cases, a virtual soft block switch integrated into the Xen HV layer maps virtual IO requests to either physical block requests serviced by the hardware driver domain, or to Ethernet frames generated by the network block protocol logic.

References

Design of the Microvisor HV architecture

The new hypervisor implementation boots as a standalone image without loading any control domain. Various major changes are required to complete the implementation of the proposed vision. The first step is the migration of the Xenbus functionality to the Xen Hypervisor layer by implementing a minimal set of Xenbus functionality inside the core Xen HV. Next, we implemented a simple key/value store (Xenstore) interface to communicate values to guest domains via Xenstore. Third, and most importantly the netback handling was integrated directly into a fast packet soft switching layer integrated directly in the Xen Hypervisor.

[1] A. Kivity, Y. Kamay, D. Laor, U. Lublin, and A. Liguori, “KVM: The Linux Virtual Machine Monitor,” in Proceedings of the Linux Symposium. [2] P. Barham, B. Dragovic, K. Fraser, S. Hand, T. Harris, A. Ho, and R. Neugebauer, “Xen and the art of virtualization,” in Proc. of the nineteenth ACM symposium on Operating systems principles (SOSP19). [3] Y. Durand, P. M. Carpenter, S. Adami, A. Bilas, D. Dutoit, A. Farcy, G. Gaydadjiev, J. Goodacre, M. Katevenis, M. Marazakis, et al., “Euroserver: Energy efficient node for european micro-servers,” in Digital System Design (DSD), 2014 17th Euromicro Conference on.

Microvisor: A Scalable Hypervisor Architecture for ...

of CPU cores, while maintaining a high performance-to- power ratio, which is the key metric if hardware is to con- tinue to scale to meet the expected demand of ExaScale computing and Cloud growth. In contrast to the KVM Hypervisor platform, the Xen. Hypervisor provides a true Type I Hypervisor. That is to say that the ...

153KB Sizes 2 Downloads 423 Views

Recommend Documents

A scalable service-oriented architecture for multimedia ... - Uni Marburg
development and use of web services for a broad scope of multimedia .... computational cloud, such as the Amazon Elastic Compute Cloud (EC2) (Amazon Web ... The service provider is interested in efficient services in that the response time.

A Scalable FPGA Architecture for Nonnegative Least ...
Each of these areas can benefit from high performance imple- mentations .... several disadvantages are immediately apparent—for each iter- ation there are two ...

A scalable service-oriented architecture for multimedia ... - Uni Marburg
BPEL for data-intensive applications and (b) provides tools that further ease the ... management and cluster/grid computing, including multidisciplinary analysis ..... Step 3 After processing the whole input video in this manner, a final list is give

FaSTAPTM: A Scalable Anti-Jam Architecture for GPS
selectable by software, but larger tap lengths can be ac- commodated with minor .... receiver does not need to be integrated with the FaSTAP. AJ module and the ...

A Wire-Delay Scalable Microprocessor Architecture for ...
technology challenges of the coming decade. The Grid Processor architecture (GPA) is designed to address these technology challenges [2]. As shown in Figure. 9.6.3, each GPA implementation consists of a 2-D array (4x4 in this example but scalable to

A Scalable Hierarchical Power Control Architecture for ...
1. SHIP: A Scalable Hierarchical Power Control. Architecture for Large-Scale Data Centers. Xiaorui Wang ... Power consumed by computer servers has become a serious concern in the ... years as more data centers reach their power limits. Therefore, ...

Jiminy: A scalable incentive-based architecture for ...
teracts with a MySQL backend database for storage and retrieval of system/user data. ..... This is part of the reason why files in file-sharing peer-to-peer systems.

Optimal Scalable Software Architecture for Symmetric Multi-Core ...
Optimal Scalable Software Architecture for Symmetric Multi-Core Embedded System.pdf. Optimal Scalable Software Architecture for Symmetric Multi-Core ...

Scalable VLSI Architecture for Variable Block Size ...
ment of Industry Science and Technology, Kitakyushu-shi, 808-. 0135 Japan. a) E-mail: ...... China in 2001 and M.E. degree in Computer. Science from Tsinghua ...

A Scalable Security Architecture Enabling Coalition ...
Abstract- Coalitions between autonomous domains are often formed in real life scenarios in order to enable access permissions to shared objects on grounds of ...

A Scalable Security Architecture Enabling Coalition ...
The dynamic nature of coalitions poses new challenges relative to security ... this paper we introduce a robust and scalable solution that enables the realization of coalition .... Multi-domain coalitions are prominent in a big number of emerging ...

Developing a Scalable Predictive-Maintenance Architecture White Paper
downtime due to tool failure, decrease overall maintenance costs, and improve productivity. ... data volumes across the network, defining baselines for FFUs, alerting ... our unscheduled downtime due to FFU failure by 300 percent over manual .... eac

Scalable VLSI Architecture for Variable Block Size ...
adopts many new features, which include variable block sizes motion .... For block size of 16 X 8, the MVPs of the upper block ... because the overlapped area of two adjacent search windows ...... phone Corporation (NTT) in 1990, where he.

A Scalable Security Architecture Enabling Coalition ...
relative to security management and joint administration of resources; therefore ... between interconnected medical information systems; e-. Government is also ...

Developing a Scalable Predictive-Maintenance Architecture White Paper
Executive Overview. To gain and maintain a competitive advantage in today's world, manufacturers from all industries are turning to Industrial Internet of Things ...

Developing a Scalable Predictive-Maintenance Architecture White Paper
are exceptional for demonstrating the ROI potential of an edge-computing ... upcoming FFU failure, the FFU is now ordered days ahead and we can plan for ... We installed sensors on each fan filter unit (FFU) to measure vibration and transmit ...

Scalable Elastic Systems Architecture
Feb 24, 2011 - vice and the cloud does not provide good primitives for the degree of elasticity you require, ... Free or idle resources become hid- den. If however, each .... of a principal. The top layer is the Elastic Service and Runtime. Layer.

A scalable architecture for use in an over-HDTV real ...
address this problem, a codec architecture using multiple encoders and decoders is an effective approach for ..... Using this system, we were able to transmit the semi nal game of the 2002 FIFA World Cup Soccer ... T.Yoshitome,K.Nakamura,K.Nitta,M.Ik

Hypervisor Support for Identifying Covertly Executing ...
based system that detects and identifies covertly execut- .... tampering with files on disk by injecting malicious code ...... worthy version of the UNIX kill utility.