Rich Wolski

Evaluating the Performance Impact of Xen on MPI and Process Execution For HPC Systems

Virtualization has become increasingly popular for enabling full system isolation, load balancing, and hardware multiplexing for high-end server systems. Virtualizing software has the potential to benefit HPC systems similarly by facilitating efficient cluster management, application isolation, full-system customization, and process migration. However, virtualizing software is not currently employed in HPC environments due to its perceived overhead. In this work, we investigate the overhead imposed by the popular, open-source, Xen virtualization system, on performance-critical HPC kernels and applications. We empirically evaluate the impact of Xen on both communication and computation and compare its use to that of a customized kernel using HPC cluster resources at Lawrence Livermore National Lab (LLNL). We also employ statistically sound methods to compare the performance of a para virtualized kernel against three popular Linux operating systems: RedHat Enterprise 4 (RHEL4) for build versions 2.6.9 and 2.6.12 and the LLNL CHAOS kernel, a specialized version of RHEL4. Our results indicate that Xen is very efficient and practical for HPC systems.

QBETS: queue bounds estimation from time series

Most space-sharing parallel computers presently operated by high-performance computing centers use batch-queuing systems to manage processor allocation. Because these machines are typically “space-shared,” each job must wait in a queue until sufficient processor resources become available to service it. In production computing settings, the queuing delay (experienced by users as the time between when the job is submitted and when it begins execution) is highly variable. Users often find this variability a drag on productivity as it makes planning difficult and intellectual continuity hard to maintain.

Eucalyptus: Delivering a Private Cloud

In an interview conducted by Dejan Milojicic, EIC of Computing Now (www.computer.org/portal/web/computingnow), Rich Wolski, CTO of Eucalyptus Systems, gives an overview of the services and applications his company provides. The extended transcript of that conversation, along with the audio podcast, is available through Computing Now.

Response time service level agreements for cloud-hosted web applications

Cloud computing is a successful model for hosting web-facing applications that are accessed by their users as services. While clouds currently offer Service Level Agreements (SLAs) containing guarantees of availability, they do not make performance guarantees for deployed applications. In this work we present Cerebro -- a system for establishing statistical guarantees of application response time in cloud settings. Cerebro combines off-line static analysis of application control structure with on-line cloud performance monitoring and statistical forecasting to predict bounds on the response time of web-facing application programming interfaces (APIs). Because Cerebro does not require application instrumentation or per-application cloud benchmarking, it does not impose any runtime overhead, and is suitable for use at cloud scales. Also, because the bounds are statistical, they are appropriate for use as the basis for SLAs between cloud-hosted applications and their users. We investigate the correctness of Cerebro predictions, the tightness of their bounds, and the duration over which the bounds persist in both Google App Engine and AppScale (public and private cloud platforms respectively). We also detail the effectiveness of our SLA prediction methodology compared to other performance bound estimation methods based on simple statistical analysis.

Performance information services for computational Grids

Grid schedulers or resource allocators (whether they be human or automatic scheduling programs) must choose the right combination of resources from the available resource pool while the performance and availability characteristics of the individual resources within the pool change from moment to moment. Moreover, the scheduling decision for each application component must be made before the component is executed making scheduling a predictive activity. A Grid scheduler, therefore, must be able to predict what the deliverable resource performance will be for the time period in which a particular application component will eventually use the resource.In this chapter, we describe techniques for dynamically characterizing resources according to their predicted performance response to enable Grid scheduling and resource allocation. These techniques rely on three fundamental capabilities: extensible and non-intrusive performance monitoring, fast prediction models, and a flexible and high-performance reporting interface. We discuss these challenges in the context of the Network Weather Service (NWS) - an on-line performance monitoring and forecasting service developed for Grid environments. The NWS uses adaptive monitoring techniques to control intrusiveness, and non-parametric forecasting methods that are lightweight enough to generate forecasts in real-time. In addition, the service infrastructure used by the NWS is portable among all currently available Grid resources and is compatible with extant Grid middleware such as Globus, Legion, and Condor.

VM-centric snapshot deduplication for cloud data backup

Data deduplication is important for snapshot backup of virtual machines (VMs) because of excessive redundant content. Fingerprint search for source-side duplicate detection is resource intensive when the backup service for VMs is co-located with other cloud services. This paper presents the design and analysis of a fast VM-centric backup service with a tradeoff for a competitive deduplication efficiency while using small computing resources, suitable for running on a converged cloud architecture that cohosts many other services. The design consideration includes VM-centric file system block management for the increased VM snapshot availability. This paper describes an evaluation of this VM-centric scheme to assess its deduplication efficiency, resource usage, and fault tolerance.

SimGate: Full-System, Cycle-Close Simulation of the Stargate Sensor Network Intermediate Node

We present SimGate - a full-system simulator for the Stargate intermediate-level, resource-constrained, sensor network device. We empirically evaluate the accuracy and performance of the system in isolation as well as coupled with simulated Mica2 motes. Our system is functionally correct and achieves accurate cycle estimation (i.e. cycle-close). Moreover, the overhead of simulated execution is modest with respect to previously published work

Using Trustworthy Simulation to Engineer Cloud Schedulers

In recent years, researchers have contributed promising new techniques for allocating cloud resources in more robust, efficient, and ecologically sustainable ways. Unfortunately, the wide-spread use of these techniques in production systems has, to date, remained elusive. One reason for this is that the state of the art for investigating these innovations at scale often relies solely on model-driven simulation. Production-grade cloud software, however, demands certainty and precision for development and business planning that only comes from validating simulation against empirical observation. In this work, we take an alternative approach to facilitating cloud research and engineering in order to transition innovations to production deployment faster. In particular, we present a new methodology that complements existing model-driven simulation with platform-specific and statistically trustworthy results. We simulate systems at scales and on time frames that are testable, and then, based on the statistical validation of these simulations, investigate scenarios beyond those feasibly observable in practice. We demonstrate the approach by developing an energy-aware cloud scheduler and evaluating it using production and synthetic traces in faster than real time. Our results show that we can accurately simulate a production IaaS system, ease capacity planning, and expedite the reliable development of its components and extensions.

Extracting spatial information from social media in support of agricultural management decisions

Farmers face pressure to respond to unpredictable weather, the spread of pests, and other variable events on their farms. This paper proposes a framework for data aggregation from diverse sources that extracts named places impacted by events relevant to agricultural practices. Our vision is to couple natural language processing, geocoding, and existing geographic information retrieval techniques to increase the value of already-available data through aggregation, filtering, validation, and notifications, helping farmers make timely and informed decisions with greater ease.

Cost-Aware Cloud Profiling, Prediction, and Provisioning as a Service

The proliferation of cloud computing in science has democratized access to large-scale computing capacity. However, while the increasing number of cloud providers and instance types offers unparalleled flexibility it also creates barriers that prohibit easy and efficient usage. New systems are required to mask the heterogeneity in a multicloud environment and provide automated provisioning of resources. Such systems require many core capabilities, including techniques that ensure appropriate instances are provisioned, sound bidding strategies, and automated provisioning methods. Here we present SCRIMP—a service-based system that provides such capabilities, enabling application developers and users to reliably outsource the task of provisioning cloud infrastructure. We show that by understanding application requirements, predicting dynamic market conditions, and automatically provisioning infrastructure according to user-defined policies and real-time conditions that our approaches can reduce costs by an order of magnitude when using commercial clouds while also improving execution performance and efficiency.