1997 · 64 citations · 31 references
ComputingComputer EngineeringComputer ScienceDistributed SystemsOperating SystemsReal-time SystemsDistributed ComputingDistributed Operating SystemsReal-time Resource ManagementReal-time Systems ArchitectureReal-time Systems DesignResource SharingClient-server CommunicationDependabilityHardware ResourcesReal-time Kernel DesignReal-time KernelsReal-time Software ArchitectureResource Sharing In Real-time SystemsSystem Dependability
Alpha is a non‑proprietary experimental kernel that extends real‑time computing to distributed applications such as telecommunications, factory automation, and defense, tackling the asynchronous, dynamic, and non‑deterministic nature of mission‑critical distributed real‑time systems that existing real‑time or distributed OS paradigms cannot scale cost‑effectively. The authors propose a new paradigm for real‑time distributed computing that seeks to manage all resources with best‑effort scheduling based on benefit functions and to enable multiparty, peer‑structured, trans‑node computations for cooperative mission management. This paradigm directly manages resources by expressing computation completion time constraints as benefit functions and implements multiparty, peer‑structured, trans‑node computations to coordinate tasks across nodes.
Alpha is a non-proprietary experimental operating system kernel which extends the realtime domain to encompass distributed applications, such as for telecommunications, factory automation, and defense. Distributed real-time systems are inherently asynchronous, dynamic, and non-deterministic, and yet are nonetheless mission-critical.The increasing complexity and pace of these systems precludes the historical reliance solely on human operators for assuring system dependability under uncertainty. Traditional real-time OS technology is based on attempting to assert or impose determinism of not just the ends but also the means, for centralized low-level sampled-data monitoring and control, with an insufficiency of hardware resources. Conventional distributed OS technology is primarily based on two-party client/server hierarchies for explicit resource sharing in networks of autonomous users. These two technological paradigms are special cases which cannot be combined and scaled up cost-effectively to accommodate distributed real-time systems. Alpha’s new paradigm for real-time distributed computing is founded on best-effort management of all resources directly with computation completion time constraints which are expressed as benefit functions; and multiparty, peer-structured, trans-node computations for cooperative mission management.
31
Judgement under Uncertainty: Heuristics and Biases
Elizabeth C. Hirschman, Daniel Kahneman, Paul Slovic et al. · Journal of Marketing Research · 1983 · 3.9K citations
End-To-End Arguments In System Design
Jerome H. Saltzer, David P. Reed, David D. Clark · 1984 · 2.1K citations
Relativity: the special and general theory
Choice Reviews Online · 2005 · 1.5K citations
Natural Sciences, History, Physics +11