Abstract: Techniques for managing energy use of a computing deployment are provided. In one embodiment, a computer system can establish a performance model for one or more components of the computing deployment, where the performance model models a relationship between one or more tunable parameters of the one or more components and an end-to-end performance metric, and where the end-to-end performance metric reflects user-observable performance of a service provided by the computing deployment. The computer system can further execute an algorithm to determine values for the one or more tunable parameters that minimize power consumption of the one or more components, where the algorithm guarantees that the determined values will not cause the end-to-end performance metric, as calculated by the performance model, to cross a predefined threshold. The computer system can then enforce the determined values by applying changes to the one or more components.

Abstract: 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. Deployments of a distributed computing application, such as Hadoop, may be executed concurrently with a distributed database application, such as HBase, using a shared instance of a distributed filesystem, or in other cases, multiple instances of the distributed filesystem. Computing resources allocated to region server nodes executing as VMs may be isolated from compute VMs of the distributed computing application, as well as from data nodes executing as VMs of the distributed filesystem.

Abstract: Techniques for managing energy use of a computing deployment are provided. In one embodiment, a computer system can establish a performance model for one or more components of the computing deployment, where the performance model models a relationship between one or more tunable parameters of the one or more components and an end-to-end performance metric, and where the end-to-end performance metric reflects user-observable performance of a service provided by the computing deployment. The computer system can further execute an algorithm to determine values for the one or more tunable parameters that minimize power consumption of the one or more components, where the algorithm guarantees that the determined values will not cause the end-to-end performance metric, as calculated by the performance model, to cross a predefined threshold. The computer system can then enforce the determined values by applying changes to the one or more components.

Abstract: An elastic filesystem for temporary data provides storage space for virtual machines (VMs) in a distributed computing system. The filesystem redirects accesses to virtual disks in VMs to a common pool file. The system provides performance and storage efficiency at least on par with local, direct attached virtual disks, while providing a single pool of shared storage that is provisioned and managed independently of the VMs. The system provides storage isolation between VMs storing temporary data in that shared pool. Also, storage space for temporary data may be allocated on demand and reclaimed when no longer needed, thereby supporting a wide variety of temporary space requirements for different Hadoop jobs.

Abstract: 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.

Abstract: In a computer-implemented method for configuring flash cache for input/output operations to a storage device by a plurality of virtual machines an input/output trace log for each of a plurality of virtual machines is accessed. Performance of each of the plurality of virtual machines based on a plurality of configurations of the flash cache is simulated in real-time. A recommendation of the plurality of configurations of the flash cache for the each of the plurality of virtual machines utilizing results from the simulation is generated.

Abstract: Exemplary methods, apparatuses, and systems receive characteristics of a plurality of input/output (I/O) requests from a workload, including logical address distance values between I/O requests and data lengths of the plurality of I/O requests. Based upon the characteristics, a data length value representative of the data lengths of the plurality of I/O requests is determined and an access pattern of the plurality of I/O requests is determined. A notification that the first workload is suitable for a virtual storage area network environment is generated based upon the characteristics. The first workload is selected as suitable in response to determining the data length value for the data lengths of the plurality of I/O requests is less than a data length threshold and/or the access pattern of the plurality of I/O requests is more random than an access pattern threshold on the spectrum from random access to sequential access.

Abstract: Exemplary methods, apparatuses, and systems receive a first input/output (I/O) trace including storage addresses that were subject to a plurality of I/O requests from a first workload during a first period of time. The first I/O trace is run through a cache simulation using a plurality of simulated cache sizes. A first state of the cache simulation is stored upon completing the first I/O trace simulation. The first I/O trace is deleted in response to storing the first state. A second I/O trace including storage addresses that were subject to a plurality of I/O requests from the first workload during a second period of time is received. A cumulative miss ratio curve for the first workload is generated by loading the stored first state as a starting point for simulating the second I/O trace and running the second I/O trace through the cache simulation.

Abstract: Exemplary methods, apparatuses, and systems that can intelligently copy data to a plurality of datastores using performance monitoring are described. In one embodiment, a shortest path tree determines the most efficient paths available for copying data from a source datastore to one or more destination datastores. During the copying of the data between a source datastore and the one or more destination datastores, a performance value of each of the datastores involved in the copying process is compared to a threshold. In response to determining that the performance value of a given source or destination datastore involved in the copying exceeds the threshold, the copying of the data to the corresponding destination datastore is suspended. An updated shortest path tree is determined to locate a more efficient path for copying data to the suspended destination datastore. Copying is resumed to the suspended destination datastore using the updated shortest path tree.

Abstract: Methods, systems, and computer programs are provided for measuring the performance of display images received on a remote computer display. One method includes an operation for detecting calls from an application to an application programming interface (API), which is provided for rendering images on a display image, each call causing an update of the display image. Further, the method includes an operation for embedding data for measuring performance in display frames of the display image based on the detecting. The embedding results in modified displayed frames with respective data for measuring performance. The modified displayed frames are transmitted to a remote client, which results in received modified display frames having respective received data for measuring the performance. In addition, the method includes an operation for calculating the remote display quality for the given application based on the received modified display frames and the respective received data for measuring performance.

Abstract: Exemplary methods, apparatuses, and systems receive a first input/output (I/O) trace from a first workload and run the first I/O trace through a cache simulation to determine a first miss ratio curve (MRC) for the first workload. A second I/O trace from the first workload is received and run through the cache simulation to determine a second MRC for the first workload. First and second cache sizes corresponding to a target miss rate for the first workload are determined using the first and second MRCs. A fingerprint of each of the first and I/O traces is generated. The first cache size, the second cache size, or a combination of the first and second cache sizes is selected as a cache size for the first workload based upon a comparison of the first and second fingerprints. A recommended cache size is generated based upon the selected cache size.

Abstract: Exemplary methods, apparatuses, and systems include a host computer selecting a first workload of a plurality of workloads running on the host computer to be subjected to an input/output (I/O) trace. The host computer determines whether to generate the I/O trace for the first workload for a first length of time or for a second length of time. The first length of time is shorter than the second length of time. The determination is based upon runtime history for the first workload, I/O trace history for the first workload, and/or workload type of the first workload. The host computer generates the I/O trace of the first workload for the selected length of time.

Abstract: Methods, systems, and computer programs are provided for measuring the performance of display images received on a remote computer display. One method includes an operation for detecting calls from an application to an application programming interface (API), which is provided for rendering images on a display image, each call causing an update of the display image. Further, the method includes an operation for embedding data for measuring performance in display frames of the display image based on the detecting. The embedding results in modified displayed frames with respective data for measuring performance. The modified displayed frames are transmitted to a remote client, which results in received modified display frames having respective received data for measuring the performance. In addition, the method includes an operation for calculating the remote display quality for the given application based on the received modified display frames and the respective received data for measuring performance.

Abstract: Exemplary methods, apparatuses, and systems that can intelligently copy data to a plurality of datastores using performance monitoring are described. In one embodiment, a shortest path tree determines the most efficient paths available for copying data from a source datastore to one or more destination datastores. During the copying of the data between a source datastore and the one or more destination datastores, a performance value of each of the datastores involved in the copying process is compared to a threshold. In response to determining that the performance value of a given source or destination datastore involved in the copying exceeds the threshold, the copying of the data to the corresponding destination datastore is suspended. An updated shortest path tree is determined to locate a more efficient path for copying data to the suspended destination datastore. Copying is resumed to the suspended destination datastore using the updated shortest path tree.

Abstract: Techniques for managing energy use of a computing deployment are provided. In one embodiment, a computer system can establish a performance model for one or more components of the computing deployment, where the performance model models a relationship between one or more tunable parameters of the one or more components and an end-to-end performance metric, and where the end-to-end performance metric reflects user-observable performance of a service provided by the computing deployment. The computer system can further execute an algorithm to determine values for the one or more tunable parameters that minimize power consumption of the one or more components, where the algorithm guarantees that the determined values will not cause the end-to-end performance metric, as calculated by the performance model, to cross a predefined threshold. The computer system can then enforce the determined values by applying changes to the one or more components.

Abstract: Exemplary methods, apparatuses, and systems receive characteristics of a plurality of input/output (I/O) requests from a workload, including logical address distance values between I/O requests and data lengths of the plurality of I/O requests. Based upon the characteristics, a data length value representative of the data lengths of the plurality of I/O requests is determined and an access pattern of the plurality of I/O requests is determined. A notification that the first workload is suitable for a virtual storage area network environment is generated based upon the characteristics. The first workload is selected as suitable in response to determining the data length value for the data lengths of the plurality of I/O requests is less than a data length threshold and/or the access pattern of the plurality of I/O requests is more random than an access pattern threshold on the spectrum from random access to sequential access.

Abstract: Exemplary methods, apparatuses, and systems receive a first input/output (I/O) trace from a first workload and run the first I/O trace through a cache simulation to determine a first miss ratio curve (MRC) for the first workload. A second I/O trace from the first workload is received and run through the cache simulation to determine a second MRC for the first workload. First and second cache sizes corresponding to a target miss rate for the first workload are determined using the first and second MRCs. A fingerprint of each of the first and I/O traces is generated. The first cache size, the second cache size, or a combination of the first and second cache sizes is selected as a cache size for the first workload based upon a comparison of the first and second fingerprints. A recommended cache size is generated based upon the selected cache size.

Abstract: Exemplary methods, apparatuses, and systems receive a first input/output (I/O) trace including storage addresses that were subject to a plurality of I/O requests from a first workload during a first period of time. The first I/O trace is run through a cache simulation using a plurality of simulated cache sizes. A first state of the cache simulation is stored upon completing the first I/O trace simulation. The first I/O trace is deleted in response to storing the first state. A second I/O trace including storage addresses that were subject to a plurality of I/O requests from the first workload during a second period of time is received. A cumulative miss ratio curve for the first workload is generated by loading the stored first state as a starting point for simulating the second I/O trace and running the second I/O trace through the cache simulation.

Abstract: Exemplary methods, apparatuses, and systems include a host computer selecting a first workload of a plurality of workloads running on the host computer to be subjected to an input/output (I/O) trace. The host computer determines whether to generate the I/O trace for the first workload for a first length of time or for a second length of time. The first length of time is shorter than the second length of time. The determination is based upon runtime history for the first workload, I/O trace history for the first workload, and/or workload type of the first workload. The host computer generates the I/O trace of the first workload for the selected length of time.

Abstract: Methods, systems, and computer programs are provided for measuring the performance of display images received on a remote computer display. One method includes an operation for detecting calls from an application to an application programming interface (API), which is provided for rendering images on a display image, each call causing an update of the display image. Further, the method includes an operation for embedding data for measuring performance in display frames of the display image based on the detecting. The embedding results in modified displayed frames with respective data for measuring performance. The modified displayed frames are transmitted to a remote client, which results in received modified display frames having respective received data for measuring the performance. In addition, the method includes an operation for calculating the remote display quality for the given application based on the received modified display frames and the respective received data for measuring performance.