Nilabja Roy

Efficient Autoscaling in the Cloud Using Predictive Models for Workload Forecasting

Large-scale component-based enterprise applications that leverage Cloud resources expect Quality of Service(QoS) guarantees in accordance with service level agreements between the customer and service providers. In the context of Cloud computing, auto scaling mechanisms hold the promise of assuring QoS properties to the applications while simultaneously making efficient use of resources and keeping operational costs low for the service providers. Despite the perceived advantages of auto scaling, realizing the full potential of auto scaling is hard due to multiple challenges stemming from the need to precisely estimate resource usage in the face of significant variability in client workload patterns. This paper makes three contributions to overcome the general lack of effective techniques for workload forecasting and optimal resource allocation. First, it discusses the challenges involved in auto scaling in the cloud. Second, it develops a model-predictive algorithm for workload forecasting that is used for resource auto scaling. Finally, empirical results are provided that demonstrate that resources can be allocated and deal located by our algorithm in a way that satisfies both the application QoS while keeping operational costs low.

A capacity planning process for performance assurance of component-based distributed systems

For service providers of multi-tiered component-based applications, such as web portals, assuring high performance and availability to their customers without impacting revenue requires effective and careful capacity planning that aims at minimizing the number of resources, and utilizing them efficiently while simultaneously supporting a large customer base and meeting their service level agreements. This paper presents a novel, hybrid capacity planning process that results from a systematic blending of 1) analytical modeling, where traditional modeling techniques are enhanced to overcome their limitations in providing accurate performance estimates; 2) profile-based techniques, which determine performance profiles of individual software components for use in resource allocation and balancing resource usage; and 3) allocation heuristics that determine minimum number of resources to allocate software components. Our results illustrate that using our technique, performance (i.e., bounded response time) can be assured while reducing operating costs by using 25% less resources and increasing revenues by handling 20% more clients compared to traditional approaches.

Design and performance evaluation of an adaptive resource management framework for distributed real-time and embedded systems

Achieving end-to-end quality of service (QoS) in distributed real-time embedded (DRE) systems require QoS support and enforcement from their underlying operating platforms that integrates many real-time capabilities, such as QoS-enabled network protocols, real-time operating system scheduling mechanisms and policies, and real-time middleware services. As standards-based quality of service (QoS) enabled component middleware automates integration and configuration activities, it is increasingly being used as a platform for developing open DRE systems that execute in environments where operational conditions, input workload, and resource availability cannot be characterized accurately a priori. Although QoS-enabled component middleware offers many desirable features, however, it historically lacked the ability to allocate resources efficiently and enable the system to adapt to fluctuations in input workload, resource availability, and operating conditions. This paper presents three contributions to research on adaptive resource management for component-based open DRE systems. First, we describe the structure and functionality of the resource allocation and control engine (RACE), which is an open-source adaptive resource management framework built atop standards-based QoS-enabled component middleware. Second, we demonstrate and evaluate the effectiveness of RACE in the context of a representative open DRE system: NASAs magnetospheric multiscale mission system. Third, we present an empirical evaluation of RACEs scalability as the number of nodes and applications in a DRE system grows. Our results show that RACE is a scalable adaptive resource management framework and yields a predictable and high-performance system, even in the face of changing operational conditions and input workload.

Toward Effective Multi-Capacity Resource Allocation in Distributed Real-Time and Embedded Systems

Effective resource management for distributed real-time embedded (DRE) systems is hard due to their unique characteristics, including (1) constraints in multiple resources and (2) highly fluctuating resource availability and input workload. DRE systems can benefit from a middleware framework that enables adaptive resource management algorithms to ensure application QoS requirements are met. This paper identifies key challenges in designing and extending resource allocation algorithms for DRE systems. We present an empirical study of bin-packing algorithms enhanced to meet these challenges. Our analysis identifies input application patterns that help generate appropriate heuristics for using these algorithms effectively in DRE systems.

Modeling Software Contention Using Colored Petri Nets

Commercial servers, such as database or application servers, often attempt to improve performance via multi-threading. Improper multi-threading architectures can incur contention, limiting performance improvements. Contention occurs primarily at two levels: (1) blocking on locks shared between threads at the software level and (2) contending for physical resources (such as the cpu or disk) at the hardware level. Given a set of hardware resources and an application design, there is an optimal number of threads that maximizes performance. This paper describes a novel technique we developed to select the optimal number of threads of a target-tracking application using a simulation-based colored Petri nets (CPNs) model. This paper makes two contributions to the performance analysis of multi-threaded applications. First, the paper presents an approach for calibrating a simulation model using training set data to reflect actual performance parameters accurately. Second, the model predictions are validated empirically against the actual application performance and the predicted data is used to compute the optimal configuration of threads in an application to achieve the desired performance. Our results show that predicting performance of application thread characteristics is possible and can be used to optimize performance.

Bulls-Eye – a resource provisioning service for enterprise distributed real-time and embedded systems

Middleware is increasingly used to develop and deploy compo nents in enterprise distributed real-time and embed ded (DRE) systems A key chal lenge in these systems is de vising resource management algorithms that deploy appli cation components properly onto target nodes To provide an accurate view of system re source utilization, these algorithms need monitor resources at runtime Runtime resource monitoring is also needed to make redeployment or reconfigu ration decisions trig gered by various factors, such as failures, attacks, overloads, or changes in quality of service (QoS) re quirements DRE sys tems with a diverse range of applications can therefore benefit from a common re source provisioning service capable of monitoring re source data and ena bling proper resource allocation in a timely manner. This paper provides two contributions to the study of run time resource provi sioning for enterprise DRE systems First, it describes the challenges in devel oping Bulls-Eye, which is an open implementation of the OMG standard Target Manager specification that provides a reusable service for provisioning distrib uted resources in enter prise DRE systems Second, it presents the results of ex periments that applied Bulls-Eye to the multi-layer resource manage ment sub system of a ship board computing environment Our re sults show that provi sioning re sources at runtime in a DRE system via Bulls-Eye simplifies resource manage ment and helps automate adaptations in the face of dynamic changes in operat ing conditions.

Impediments to Analytical Modeling of Multi-Tiered Web Applications

Service providers hosting multi-tiered applications require accurate analytical models of the applications they will host for different system management activities, such as capacity planning, configuration management, cost analysis and feedback control. Due to the complexity of real world scenarios, developing accurate analytical models is hard. This paper presents the commonly faced challenges in developing these analytical models that stem from real-world issues, such as excessive system activity, presence of multiple cores or processors, and concurrency management. Presence of multi-tiered applications further compounds the challenges faced. We sketch preliminary ideas based on application-specific and/or domain-specific modeling techniques to overcome these limitations.

A Component Assignment Framework for Improved Capacity and Assured Performance in Web Portals

Web portals hosting large-scale internet applications have become popular due to the variety of services they provide to their users. These portals are developed using component technologies. Important design challenges for developers of web portals involve (1) determining the component placement that maximizes the number of users/requests (capacity) without increasing hardware resources and (2) maintaining the performance within certain bounds given by service level agreements (SLAs). The multitude of behavioral patterns presented by users makes it hard to identify the incoming workloads. This paper makes three contributions to the design and evaluation of web portals that address these design challenges. First it introduces an algorithmic framework that combines bin-packing and modeling-based queuing theory to place components onto hardware nodes. This capability is realized by the Component Assignment Framework for multi-tiered internet applications (CAFe). Second, it develops a component-aware queuing model to predict web portal performance. Third, it provides extensive experimental evaluation using the Rice University Bidding System (RUBiS). The results indicate that CAFe can identify opportunities to increase web portal capacity by 25% for a constant amount of hardware resources and typical web application and user workloads.

The Impact of Variability on Soft Real-Time System Scheduling

Soft real-time systems sometimes operate under uncertain and unpredictable environmental conditions which makes event arrival times unreliable and variable. Input to such systems also change from time to time making event processing times variable. Due to such variations, traditional techniques using worst case times to estimate system performance deviate far from actual expected behavior.This paper presents a Method of Stages based Analysis of soft Real Time systems (MoSART). MoSART takes into account variance in both the arrival and execution time and can model the performance of different scheduling algorithms. Sensitivity analysis, experimental validation, and the discovery of state dependent algorithms that outperform popular algorithms are demonstrated.

Abstract only: a capacity planning process for performance assurance of component-based distributed systems

For service providers of multi-tiered component-based applications, such as web portals, assuring high performance and availability to their customers without impacting revenue requires effective and careful capacity planning that aims at minimizing the number of resources, and utilizing them efficiently while simultaneously supporting a large customer base and meeting their service level agreements. This paper presents a novel, hybrid capacity planning process that results from a systematic blending of 1) analytical modeling, where traditional modeling techniques are enhanced to overcome their limitations in providing accurate performance estimates; 2) profile-based techniques, which determine performance profiles of individual software components for use in resource allocation and balancing resource usage; and 3) allocation heuristics that determine minimum number of resources to allocate software components. Our results illustrate that using our technique, performance (i.e., bounded response time) can be assured while reducing operating costs by using 25% less resources and increasing revenues by handling 20% more clients compared to traditional approaches.