Component-level virtual machines: A paradigm for achieving controllability in cloud-deployed dynamic distributed systems

点击翻译按钮获取中文摘要

Component-based software engineering (CSBE) has become standard, e.g., Microsoft's COM, Sun's EJB, and CORBA CCM. Modern societies have become reliant on CSBE systems for core services across many diverse domains, e.g., entertainment and social networking, banking and finance, healthcare, eGovernment, and critical infrastructure. But, ensuring that real-world CSBE systems are robust to failure, meet their quality of service (QoS) guarantees, etc., has remained an open problem, in part due to the lack of predictability of server-level behaviours. Queuing theory provide abstracted models of CSBE system behaviours that tend to become suspect towards a system's performance, capacity, and fault boundaries and neglect system dynamics and transients. Cloud-deployed dynamic distributed systems (DDS) offers a paradigm to address some of these issues by enabling distributed systems that grow/shrink in real-time in response to workload demand. But, DDS introduce closed-loop control theory issues into software systems given DDS use feed back control loops, i.e., stability, transients, potential oscillatory behaviours, etc., all come into play. This research program will address the problem of controllability in cloud-deployed DDS systems through introducing component-level virtual machines (CVMs) as a pragmatic approach that wraps each CSBE component its own minimally scoped VM, with full CSBE systems then composed via CVM-to-CVM message passing. This places component behaviours into hard VM-enforced shells thereby allowing: a) CSBE system predictability, b) tractable control theory analyzes, c) a practical method for applying formal method constructs via CVM-enforced message policies, and d) potentially, significantly improving cloud resource utilization over their current 20% range, a core facet to improving carbon footprints within clouds. Overall, the research will significantly advance the state of the art in building at-scale software systems that meet QoS guarantees, in a manner that is usable by real-world programmers.