Virtualization can provide significant benefits in data centers by enabling virtual machine migration to eliminate hotspots. We present Sandpiper, a system that automates the task of monitoring and detecting hotspots, determining a new mapping of physical to virtual resources and initiating the necessary migrations. Sandpiper implements a black-box approach that is fully OS- and application-agnostic and a gray-box approach that exploits OS- and application-level statistics. We implement our techniques in Xen and conduct a detailed evaluation using a mix of CPU, network and memory-intensive applications. Our results show that Sandpiper is able to resolve single server hotspots within 20 seconds and scales well to larger, data center environments. We also show that the gray-box approach can help Sandpiper make more informed decisions, particularly in response to memory pressure.

/pdf/black-box-and-gray-box-strategies-for-virtual-machine-4dlglfch88.pdf

Black-box and gray-box strategies for virtual machine migration

Cloud computing allows business customers to scale up and down their resource usage based on needs. Many of the touted gains in the cloud model come from resource multiplexing through virtualization technology. In this paper, we present a system that uses virtualization technology to allocate data center resources dynamically based on application demands and support green computing by optimizing the number of servers in use. We introduce the concept of "skewness&#x201D; to measure the unevenness in the multidimensional resource utilization of a server. By minimizing skewness, we can combine different types of workloads nicely and improve the overall utilization of server resources. We develop a set of heuristics that prevent overload in the system effectively while saving energy used. Trace driven simulation and experiment results demonstrate that our algorithm achieves good performance.

Dynamic Resource Allocation Using Virtual Machines for Cloud Computing Environment

A management capability is provided for a virtual computing platform. In one example, this platform allows interconnected physical resources such as processors, memory, network interfaces and storage interfaces to be abstracted and mapped to virtual resources (e.g., virtual mainframes, virtual partitions). Virtual resources contained in a virtual partition can be assembled into virtual servers that execute a guest operating system (e.g., Linux). In one example, the abstraction is unique in that any resource is available to any virtual server regardless of the physical boundaries that separate the resources. For example, any number of physical processors or any amount of physical memory can be used by a virtual server even if these resources span different nodes. A virtual computing platform is provided that allows for the creation, deletion, modification, control (e.g., start, stop, suspend, resume) and status (i.e., events) off the virtual servers which execute on the virtual computing platform and the management capability provides controls for these functions. In a particular example, such a platform allows the number and type of virtual resources consumed by a virtual server to be scaled up or down when the virtual server is running. For instance, an administrator may scale a virtual server manually or may define one or more policies that automatically scale a virtual server. Further, using the management API, a virtual server can monitor itself and can scale itself up or down depending on its need for processing, memory and I/O resources. For example, a virtual server may monitor its CPU utilization and invoke controls through the management API to allocate a new processor for itself when its utilization exceeds a specific threshold. Conversely, a virtual server may scale down its processor count when its utilization falls. Policies can be used to execute one or more management controls. More specifically, a management capability is provided that allows policies to be defined using management object's properties, events and/or method results. A management policy may also incorporate external data (e.g., an external event) in its definition. A policy may be triggered, causing the management server or other computing entity to execute an action. An action may utilize one or more management controls. In addition, an action may access external capabilities such as sending notification e-mail or sending a text message to a telephone paging system. Further, management capability controls may be executed using a discrete transaction referred to as a 'job.' A series of management controls may be assembled into a job using one or management interfaces. Errors that occur when a job is executed may cause the job to be rolled back, allowing affected virtual servers to return to their original state.

System and method for managing virtual servers

Power delivery, electricity consumption, and heat management are becoming key challenges in data center environments. Several past solutions have individually evaluated different techniques to address separate aspects of this problem, in hardware and software, and at local and global levels. Unfortunately, there has been no corresponding work on coordinating all these solutions. In the absence of such coordination, these solutions are likely to interfere with one another, in unpredictable (and potentially dangerous) ways. This paper seeks to address this problem. We make two key contributions. First, we propose and validate a power management solution that coordinates different individual approaches. Using simulations based on 180 server traces from nine different real-world enterprises, we demonstrate the correctness, stability, and efficiency advantages of our solution. Second, using our unified architecture as the base, we perform a detailed quantitative sensitivity analysis and draw conclusions about the impact of different architectures, implementations, workloads, and system design choices.

https://people.cs.pitt.edu/~kirk/cs3150spring2010/2008_asplos_nopowerstruggles.pdf

No "power" struggles: coordinated multi-level power management for the data center

A cloud computing environment having a plurality of computing nodes is described. A plurality of permissions associated with a cloud customer is created. A first set of permissions from the plurality of permissions is associated with one or more objects. Each of the first set of permissions describes an action performed on an object. A second set of permissions from the plurality of permissions is associated with one or more users. Each of the second set of permissions describes an action to be performed by one or more users.

Virtual computing infrastructure

This paper describes the design and implementation of a system that uses virtual machine technology [1] to provide fast, transparent application migration. This is the first system that can migrate unmodified applications on unmodified mainstream Intel x86-based operating system, including Microsoft Windows, Linux, Novell NetWare and others. Neither the application nor any clients communicating with the application can tell that the application has been migrated. Experimental measurements show that for a variety of workloads, application downtime caused by migration is less than a second.

/pdf/fast-transparent-migration-for-virtual-machines-2zsc1b2yr1.pdf

Fast transparent migration for virtual machines

A source virtual machine (VM) hosted on a source server is migrated to a destination VM on a destination server without first powering down the source VM. After optional pre-copying of the source VM's memory to the destination VM, the source VM is suspended and its non-memory state is transferred to the destination VM; the destination VM is then resumed from the transferred state. The source VM memory is either paged in to the destination VM on demand, or is transferred asynchronously by pre-copying and write-protecting the source VM memory, and then later transferring only the modified pages after the destination VM is resumed. The source and destination servers preferably share common storage, in which the source VM's virtual disk is stored; this avoids the need to transfer the virtual disk contents. Network connectivity is preferably also made transparent to the user by arranging the servers on a common subnet, with virtual network connection addresses generated from a common name space of physical addresses.

Virtual machine migration

A computer architecture includes a first operating system (COS), which may be a commodity operating system, and a kernel, which acts as a second operating system. The COS is used to boot the system as a whole. After booting, the kernel is loaded and displaces the COS from the system level, meaning that the kernel itself directly accesses predetermined physical resources of the computer. All requests for use of system resources then pass via the kernel. System resources are divided into those that, in order to maximize speed, are controlled exclusively by the kernel, those that the kernel allows the COS to handle exclusively, and those for which control is shared by the kernel and COS. In the preferred embodiment of the invention, at least one virtual machine (VM) runs via a virtual machine monitor, which is installed to run on the kernel. Each VM, the COS, and even each processor in a multiprocessor embodiment, are treated as separately schedulable entities that are scheduled by the kernel. Mechanisms for high-speed I/O between VM's and I/O devices are also included.

Computer configuration for resource management in systems including a virtual machine

A distributed computing application is described that provides a highly elastic and multi-tenant platform for Hadoop applications and other workloads running in a virtualized environment. Data and compute nodes are separated into different virtual machines (VM). Compute VMs are used to launch containers from different tenants. Compute VMs are organized in pools of hot spare VMs that are immediately available for launching a container and executing a task, and pools of cold spare VMs. Each compute VM may include a mounted network filesystem provided by a node manager to share intermediate outputs across VMs executing on the same host.

Container virtual machines for hadoop

Techniques for creating a fault tolerant system in a virtual machine environment utilize a primary VM and a backup VM. To initialize the fault tolerant system, the backup VM and primary VM start from the same state. To achieve this in one embodiment, the primary VM is suspended and the state of the primary VM is copied to the backup VM. Once the backup VM has received all the primary VM's state, the primary VM is resumed. Subsequent state changes of the primary VM are buffered until the backup VM resumes, connects to the primary VM, and starts consuming the buffered content. Thereafter, synchronization is maintained by the primary VM's writing relevant state changes to a log and the backup VM's reading such relevant state changes from the log.

Michael Nelson

Papers

Fast transparent migration for virtual machines

Virtual machine migration

Computer configuration for resource management in systems including a virtual machine

Container virtual machines for hadoop

Using virtual machine cloning to create a backup virtual machine in a fault tolerant system