Adapting Distributed Real-time and Embedded Pub/SubMiddleware for Cloud Computing EnvironmentsJoe Hoffert, Douglas C. Schmidt, and Aniruddha GokhaleVanderbilt University, VU Station B #1829, 2015 Terrace Place, Nashville, TN 37203Abstract. Enterprise distributed real-timeand embedded (DRE) publish/subscribe(pub/sub) systems manage resources and data that are vital to users. Cloud com-puting—where computing resources are provisioned elastically and leased as aservice—is an increasingly popular deployment paradigm. Enterprise DRE pub/-sub systems can leverage cloud computing provisioning services to execute neededfunctionality when on-site computing resources are not available. Although cloudcomputing provides flexible on-demand computing and networking resources,enterprise DRE pub/sub systems often cannot accurately characterize their be-havior a priori for the variety of resource configurations cloud computing sup-plies (e.g., CPU and network bandwidth), which makes it hard for DRE systemsto leverage conventional cloud computing platforms.This paper provides two contributions to the study of how autonomic configu-ration of DRE pub/sub middleware can provision and use on-demand cloud re-sources effectively. We first describe how supervised machine learning can con-figure DRE pub/sub middleware services and transport protocols autonomicallyto support end-to-end quality-of-service (QoS) requirements based on cloud com-puting resources. We then present results that empirically validate how comput-ing and networking resources affect enterprise DRE pub/sub system QoS. Theseresults show how supervised machine learning can configure DRE pub/sub mid-dleware adaptively in < 10 μsec with bounded time complexity to support keyQoS reliability and latency requirements.Keywords: Autonomic configuration, pub/sub middleware, DRE systems, cloudcomputing1 IntroductionEmergingtrends and challenges. Enterprise distributed real-time and embedded (DRE)publish/subscribe (pub/sub) systems manage data and resources that are critical to theongoing system operations. Examples include testing and training of experimental air-craft across a large geographic area, air traffic management systems, and disaster recov-ery operations. These types of enterprise DRE systems must be configured correctly toleverage available resources and respond to the system deployment environment. Forexample, search and rescue missions in disaster recovery operations need to config-ure the image resolution used to detect and track survivors depending on the availableresources (e.g., computing power and network bandwidth) [20].This work is sponsored by NSF TRUST and AFRL.Contact author’s email address: [email protected] Joe Hoffert, Douglas C. Schmidt, and Aniruddha GokhaleMany enterprise DRE systems are implemented and developed for a specific com-puting/networking platform and deployed with the expectation of specific computingand networking resources being available at runtime. This approach simplifies develop-ment complexity since system developers need only focus on how the system behavesin one operating environment. Thus considerations of multiple infrastructure platformsare ameliorated with respect to system quality-of-service (QoS) properties (e.g., re-sponsiveness of computing platform, latency and reliability of networked data, etc.).Focusing on only a single operating environment, however, decreases the flexibility ofthe system and makes it hard to integrate into different operating environments, e.g.,porting to new computing and networking hardware.Cloud computing [6, 17] is an increasingly popular infrastructure paradigm wherecomputingand networkingresources are providedto a system or application as a service—typically for a “pay-as-you-go” usage fee. Provisioning services in cloud environmentsrelieve enterprise operators of many tedious tasks associated with managing hardwareand software resources used by systems and applications. Cloud computing also pro-vides enterprise application developers and operators with additional flexibility by vir-tualizing resources, such as providing virtual machines that can differ from the actualhardware machines used.Several pub/sub middleware platforms (such as the Java Message Service [16], andWeb Services Brokered Notification [14]) can (1) leverage cloud environments,(2) sup-port large-scale data-centric distributed systems, and (3) ease development and deploy-ment of these systems. These pub/sub platforms, however, do not support fine-grainedand robust QoS that are needed for enterprise DRE systems. Some large-scale dis-tributed system platforms, such as the Global Information Grid [1] and Network-centricEnterprise Services [2], require rapid response, reliability, bandwidth guarantees, scal-ability, and fault-tolerance.Conversely, conventional cloud environments are problematic for enterprise DREsystems since applicationswithin these systems often cannot characterize the utilizationof their specific resources (e.g., CPU speeds and memory) accurately a priori. Conse-quently, applications in DRE systems may need to adjust to the available resourcessupplied by the cloud environment (e.g., using compression algorithms optimized forgiven CPU power and memory) since the presence/absence of these resources affecttimeliness and other QoS properties crucial to proper operation. If these adjustmentstake too long the mission that the DRE system supports could be jeopardized.Configuring an enterprise DRE pub/sub system in a cloud environment is hard be-cause the DRE system must understand how the computing and networking resourcesaffect end-to-end QoS. For example, transport protocols provide different types of QoS(e.g., reliability and latency) that must be configured in conjunction with the pub/submiddleware. To work properly, however, QoS-enabled pub/sub middleware must un-derstand how these protocols behave with different cloud infrastructures. Likewise, themiddleware must be configured with appropriate transport protocols to support the re-quired end-to-endQoS. Manual or ad hoc configuration of the transport and middlewarecan be tedious, error-prone, and time consuming.Solution approach → Supervised Machine Learning for Autonomous Config-uration of DRE Pub/Sub Middleware in Cloud Computing Environments. ThisADAMANT 3paper describes how we are (1) evaluating multiple QoS concerns (i.e., reliability andlatency) based on differences in computing
View Full Document
Unlocking...