PARTIES

Cronus [1-7] is an object oriented distributed system which operates in a heterogeneous computer environment of interconnected local area networks. As a distributed system architecture. Cronus faces a number of resource management issues not present in non-distributed architectures. Strategies for effectively controlling the redundancy and reconfigurability inherent in Cronus are needed to take advantage of the distributed system environment. These strategies for resource management are often conveniently separated into policies and mechanisms. A policy is a goal or guideline set by a system administrator or component designer constraining the decisions made by a resource allocator. A mechanism is an internal system structure designed to implement a class of policies.In the Cronus distributed system model, there are currently two general aspects of resource allocation mechanisms which are particular to the network environment and must be effectively managed. One of these is the binding of a client request to a particular resource manager for those resources which are available redundantly. Redundancy comes in two forms: replicated objects, (e.g., a multi-copy file) and replicated managers (any manager for a type can create a new instance of that object type). For both forms the selection of an object manager to provide the given service is an important resource management policy decision. The other important aspect of resource management mechanism is the ability to reconfigure parts of the system by dynamically migrating objects. Object migration is a powerful tool for matching system resources to tasks in a manner that attempts to maximize some measure of system performance, reliability, or survivability. Both static reconfiguration (e.g., choosing or amending the placement of object managers and their associated objects), and dynamic reconfiguration (e.g., moving an individual object in direct response to demand for its use possible in the Cronus architecture and design.The general approach to resource management in Cronus is to individually control the management of the classes of objects which make up the system. This approach extends Cronus resource management concepts beyond system resources to the abstract resources developed by applications. Resource management for an individual abstract resource (type) and for groups of related resource types (services) in Cronus is based on combining a number of carefully planned mechanisms which are part of the system architecture. In addition to resource management by resource type, a client specific policy which considers collections of object types used in a specific context can be constructed.In Cronus we achieve system wide and easily controllable resource management by requiring the object managers to cooperate in enforcing a resource management policy for their resource type. An object manager can redirect operations to a peer manager on another processor on the basis of current resource status. In the case for creation of a file to an alternate file manager which may have more storage available. Part of the basis for decisions to redirect requests are parameters, settable dynamically by system administrators through monitoring and control functions, which control the resource management strategy. The creation of objects and resource management in general thus becomes a responsibility that is decentralized among object managers on each processor based on a global allocation policy, both in terms of sharing current status and possible redirection of operations between managers.The Cronus resource management model is based on the integration of the following set of primitive mechanisms: the kernel routes requests for a given type to any available manager of that type, using the Locate mechanismCronus managers redirect requests to a more appropriate peer to accomplish resource management objectivesmanagers periodically collect current status of their peers, via a standard mechanism (ReportStatus), to be used as a basis for selecting a site for redirecting an operationusers and applications optionally indicate specific location preferences with requestsan operator can monitor and regulate the functioning of the resource management policyapplications can use the basic system mechanisms (e.g. broadcast or multicast, ReportStatus operations) to build special-purpose resource management strategies tailored to their needThere is a hierarchy of desirable locations for implementing resource management policies: object managers, shared libraries, and finally client programs or users themselves. Managers are the most desirable because they are readily identifiable with a limited set of object types which are likely to have similar policy goals. Also, managers are addressable as a group for administrative control to regulate the policy parameters, and for peer manager cooperation to implement the policy. Many decisions are negotiated between managers based upon information periodically obtained via the ReportStatus operation. The Cronus Monitoring and Control System uses the same ReportStatus operation to present the operator with a global view of how well the resource management policy is proceeding.In this paper we elaborate on these approaches to distributed resource management in Cronus, including appropriate background, detailed system design, and experiences with incorporating these ideas into three different resource managers.

Recently, there has been an increased reliance on computing systems supported by a multi-tier architecture. In multi-tier computing systems, it is important to appropriately manage resource allocation to ensure fairness of a QoS (quality of service) level avoiding overload conditions in tiers. This paper proposes an adaptive resource management algorithm for multi-tier computing systems in order that all clients have the same QoS level. We introduce a computing architecture which consists of multiple tiers, a group of resource managers, and an arbiter. Each tier is specialized to execute each subtask of clients and hosts virtual machines on its server pool. Each resource manager handles resource allocation of each client and updates the resources by locally exchanging a QoS level of its client with some other resource managers. Then, the resource managers request the resources to the arbiter. The arbiter compensates the requested resources to avoid overload conditions in tiers. Based on the compensation by the arbiter, each resource manager reallocates the resources to the subtasks of its client. We show sufficient conditions for the proposed resource management algorithm to achieve a fair QoS level avoiding overload conditions in all tiers at each time.

Managers of natural resources are faced with increasingly complex issues involving resource management concerns and the emerging importance of the socio-cultural context of resource use. The resource manager has been forced to respond and react to new social, cultural, ecological, and economic realities. Educational institutions have often been slow to respond to the changing work environment, even though a growing number of practitioners and educators believe that traditional approaches to training resource managers are inadequate and increasingly, ineffectual. This paper outlines the development of a new curriculum attempting to respond to the new challenges faced by contemporary resource managers. The concepts of interdisciplinary and cross-disciplinary study are described, and the need for a new educational approach is argued. Finally, one institution's response to the changing needs and skills of resource managers is reviewed.

Chapter 8 - Two-Phase Commit

The objective of the MEMO project is to design, integrate and operate applications which utilise interactive DAB services, i.e. combining the existing GSM mobile telecommunications network and the Digital Audio Broadcasting system. A first trial with this platform for mobile multimedia services was performed to demonstrate the results achieved after one year of work and cooperation of the European partners involved. The MEMO architecture comprises DAB transmitter equipment and PC-based terminals. During the specification of the first generation prototype and the software/hardware development the following results were achieved: Infrastructure and transmitter equipment (Specification and implementation of a protocol for information interchange between contents providers and the DAB network; design and implementation of the multi-network server (MNS) with an interface to the DAB-network-server (DNS) to feed the DAB transmitter; set-up of a DAB test transmitter at CCETT, Rennes; specification of a protocol for broadcasting Multimedia data on DAB) and Terminal (development of prototype DAB receivers with a data interface and PC plug-in card, notebook PCs and interface equipment to connect the PCs to the DAB receiver for data transfer at data rates up to 256 kbit/s; implementation of the GSM interaction channel between the mobile terminal and the MNS; definition and partial implementation of Application Programming Interfaces (APIs); development of terminal software for the network/system/application service layer and adaptation of publishing and construction industry applications to demonstrate mobile, interactive services). For the trial, the transmission chain from GSM request to DAB download was successfully completed and operated, including the connection between information provider and network operator. Three interactive applications were implemented and tested — a specialised software for the building construction industry, a system to download issues of a newspaper and a general software to transmit data objects — representing the contrary edges of services MEMO is targeting. Thus, the approach of interactive mobile services at high data rates was presented for the first time to potential customers and users providing valuable results.

The work described in this paper began with a desire to include some linguistic concept of a resource manager within a dataflow language we have been designing [AGP76]. In doing so, we discovered that dataflow monitors (resource managers) provide a natural way of thinking about resources and especially their scheduling. Dataflow semantics are based upon a program composed of asynchronous operators interconnected by lines along which data tokens (messages) flow, such that when all of the input tokens for a given operator have arrived then that operator may fire (execute) by absorbing the input tokens, computing, and producing an output token as its result. These operations closely match one's intuitive model of resource managers (operators) passing signals (tokens) to one another for the purpose of synchronizing and scheduling resource usage. Previously though, dataflow languages [D73, K73, W75] have dealt only with the expression of highly asynchronous yet determinate computations; however, resource management characteristically involves indeterminate computation. The introduction here of dataflow monitors and an explicit nondeterministic merge operator for dataflow streams makes dataflow very well suited for expressing interprocess communication and operations on resources.

Generative AI, especially LLM, is driving a fundamental shift in software paradigms, prompting cloud providers to build more efficient serving infrastructures. To meet the computational demands of emerging software, modern CPU processors are integrating Accelerator Units (AU) in the pipeline to accelerate key operations, such as Intel AMX for matrix multiplication. Current practices that dedicate AU-enabled CPU exclusively to LLM serving lead to significant resource waste and inferior efficiency. To this end, sharing AU-enabled CPU with general workloads is necessary to harvest redundant resources and improve platform performance-per-watt. However, perfectly sharing AU can be challenging since they introduce three-dimensional variations: variable usage patterns, compulsory frequency interferences, and dissimilar resource bounds. Existing resource managers are oblivious to complex Accelerator Unit Variations (AUV), resulting in performance and efficiency degradations of up to 50 % in shared environments. Therefore, this paper introduces AUM, a novel AU-aware resource manager designed to handle AUV and maximize the efficiency of shared processors. AUM has two cooperative components with three stages for three-dimensional AUV. The background profiler characterizes the usage, frequency, and resource information into a discrete model, guiding the runtime controller to analyze usage-aware requirements, select frequency-aware divisions, and make bound-aware resource decisions. Through extensive evaluations on production AU-enabled CPUs, we show that AUM improves CPU efficiency by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$4.7-8.8 \%$</tex> while maintaining high-performance AU applications by reducing SLO violations by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7 - 1 1 \%}$</tex> compared with state-of-the-art resource managers.

Edge machine learning (Edge ML) is expected to serve as a key enabler for real-time mobile health (mHealth) applications. However, its reliability is governed by the limited energy and computing resources of user equipment (UE), along with the wireless channel variations and dynamic resource allocation at edge servers. In this letter, we incorporate both UE and edge server computing to satisfy the strict latency requirements of mHealth applications while efficiently utilizing the UE’s energy resources. Specifically, we separate the feature extraction and classification processes of Edge ML inference and formulate an optimization problem to distribute them between the UE and the edge server while determining the optimal UE transmit power. We demonstrate the effectiveness of the proposed approach using an mHealth case study for predicting epileptic seizures using data from wearable health devices.

Many existing MAC protocols for wireless sensor networks try to achieve simultaneously low latency and low power consumption. For many monitoring applications, however, there are two separated types of traffic. Most of the traffic is periodic, and is not subject to latency constraints: energy savings is the main objective for this traffic. A small proportion of the traffic has strict latency requirements, and the energy consumed for this priority traffic is not an issue. In this paper, we introduce a MAC protocol that can achieve this traffic differentiation in a WSN. We propose to combine the low power listening approach for periodic traffic with opportunistic encounters mechanisms for urgent traffic. We show that our approach provides a good ratio of consumed energy per delivered packet of the periodic traffic, while keeping a low loss rate for the priority traffic.

Infrared (IR) technology has emerged as a solution for monitoring dark environments. It offers resilience to shifting illumination, appearance changes, and shadows, with applications spanning self-driving cars, robotics, nighttime security, and many other fields. While existing state-of-the-art RGB-based human action recognition (AR) models exhibit limitations in scalability for action understanding under uncertain, low-light, or dark conditions. Integrating these with IR data faces challenges due to changes in modality, high resource demands, and strict latency requirements. Such issues hinder the deployment of these technologies in real-world settings. To overcome these challenges, we introduce a novel slow-fast tubelet (SFT) processing framework designed for efficient and accurate AR in IR-based scenarios. The SFT framework comprises three modules: tubelet preprocessing (TPP), feature extraction, and the feature lateral connection and recognition module (FELCM). The TPP module refines IR streams by extracting the region of interest, filtering detected objects, removing noise, and generating tubelets of refined frames. The FELCM processes refined tubelet through two pathways, where the fast tubelet path operates at a high rate and the slow tubelet path operates at a slow rate. These pathways interconnect through lateral connections, facilitating mutual updates, and enhancing the prediction efficiency. We conducted extensive experiments on benchmark datasets, including NTURGB-D 120 and infrared action recognition (InfAR). The results demonstrate that our proposed SFT framework surpasses state-of-the-art approaches in terms of accuracy (2.7% and 3.3% improvement, respectively), computational cost, and inference latency while maintaining the competitive recognition performance. Our framework's promising results underscore its potential for direct deployment in real-world applications.

SummaryLong‐running stream applications usually share the same fundamental computational infrastructure. To improve the efficiency of data processing in stream processing systems, a data analysis operator could be partitioned into n parallel tasks. The partitioned tasks are usually deployed on m nodes coexisting with other application operators. Because the node performance can vary in unpredictable ways (i.e., (1) stream input rates may fluctuate and (2) computational resource availability varies as other applications are affected), the nodes have different processing steps, and the slow node determines the operator performance. Hence, the tasks should be redistributed at runtime for stream applications to meet their strict latency requirements. Our key idea is to redistribute the tasks to the best node dynamically adaptive to resource or load fluctuations. In this paper, we present a runtime‐aware adaptive schedule mechanism that aims at minimizing the operator processing latency and minimizing the latency difference between different nodes' tasks. We propose a new abstraction called performance cost ratio (PCR) that evaluates the node performance. The higher the node's PCR is, the less cost the node will pay for processing one tuple, and the more tasks should be deployed on it. In a scheduling, we first sort tasks descendingly by their loads and sort nodes by their PCR. Then we reassign the amount of computation according to the node's PCR to keep the node's PCR and its input rate the same or in similar proportion in all PCRs. The PCR‐based quantitative algorithm applies itself to make tasks loads quantized to the processing capacity of nodes, move the minimum amount of operator's tasks, and keep the tasks local at the same time. We have implemented a runtime‐aware adaptive scheduler as an extension to Storm and evaluated this strategy. We achieve the optimization goal using less computational resources. Copyright © 2015 John Wiley &amp; Sons, Ltd.

Senckenberg’s natural history collections encompass over 45 million physical specimens distributed across 11 facilities, with 1.6 million digitized records accessible in 124 collections. Additional digital objects are stored in various infrastructures, such as Edaphobase, a digital repository for harmonized soil information (physical, chemical, and biological), and the WildlIVE Portal, a platform for FAIR (Findable, Accessible, Interoperable, and Reusable) data sharing of biodiversity monitoring with edge sensors such as camera traps. Managing this heterogeneous landscape, ranging from legacy specimen data from the pre-digital era to newly digitized objects, presents significant challenges regarding legal compliance, data sovereignty, and the implementation of FAIR principles.To address this volume, the FAIRenrich workflow automates the semantic annotation and maintenance of existing digital collection data. Handling the complexity of such enrichment requires AI models suited for partially non-deterministic tasks, incorporating an optional human-in-the-loop mechanism. By executing these workflows on a distributed network of stationary and mobile edge computing devices ('last-mile AI'), the architecture ensures strict adherence to data sovereignty and privacy requirements.Beyond data curation, FAIRenrich's distributed architecture enables systemic efficiency through resource pooling informed by industry practice. Institutional edge infrastructure is rarely fully utilized; by networking heterogeneous devices, the system dynamically reallocates idle capacity to enrichment tasks. This mirrors industry practice: major technology operators, such as Google, systematically redeploy hardware from their refresh cycles into secondary-use programs.FAIRenrich extends this model to legacy hardware: rather than treating end-of-life equipment as waste, the system enables cost-effective redeployment for delay-tolerant semantic enrichment tasks, such as inference workloads without strict latency requirements. By aligning workload scheduling to renewable peaks (e.g., photovoltaic installations), the approach implements carbon-aware scheduling principles used by major technology operators, achieving both infrastructure cost reduction and extended hardware lifecycles. This creates a circular-economy model for research institutions, transforming refresh-cycle surplus into productive scientific infrastructure.This contribution demonstrates how FAIRenrich enables sustainable semantic annotation through a distributed edge architecture that simultaneously ensures data sovereignty, optimizes infrastructure utilization, and can realize cost-effective redeployment of legacy hardware. The approach exemplifies a scalable blueprint for research institutions seeking to decouple semantic enrichment from project-resource limitations through parallelization, temporal flexibility, and circular infrastructure practices.

Distributed dataflow systems such as Apache Spark allow the execution of iterative programs at large scale on clusters. In production use, programs are often recurring and have strict latency requirements. Yet, choosing appropriate resource allocations is difficult as runtimes are dependent on hard-to-predict factors, including failures, cluster utilization and dataset characteristics. Offline runtime prediction helps to estimate resource requirements, but cannot take into account inherent variance due to, for example, changing cluster states. We present SMiPE, a system estimating the progress of iterative dataflows by matching a running job to previous executions based on similarity, capturing properties such as convergence, hardware utilization and runtime. SMiPE is not limited to a specific framework due to its black-box approach and is able to adapt to changing cluster states reflected in the current job’s statistics. SMiPE automatically adapts its similarity matching to algorithm-specific profiles by training parameters on the job history. We evaluated SMiPE with three iterative Spark jobs and nine datasets. The results show that SMiPE is effective in choosing useful historic runs and predicts runtimes with a mean relative error of 9.1% to 13.1%.

Frequency-modulated continuous-wave (FMCW) millimeter-wave (mmWave) radar systems are increasingly utilized in environmental sensing due to their high range resolution and robust sensing ability in severe weather environments. However, mutual interference among radar systems significantly degrades the target detection capability. Recent advancements in interference mitigation utilizing deep learning (DL) approaches have demonstrated promising results. DL-based approaches typically have high computational costs, which makes them unsuitable for real-time applications with strict latency requirements and limited computing resources. In this paper, we propose an efficient solution for real-time radar interference mitigation. A lightweight transformer, which is smaller and faster than the baseline transformer, is designed to reduce interference. The integration of linear attention mechanisms with depthwise separable convolutions significantly reduces the network’s computational complexity while maintaining a comparable performance. In addition, a two-stage knowledge distillation (KD) process is deployed to compress the network and enhance its efficiency. The staged distillation approach alleviates the training difficulties associated with substantial differences between the teacher and student networks. Both simulated and real-world experiments demonstrate that the proposed method outperforms the state-of-the-art methods while achieving high processing speeds.

The rapid electrification of the transportation sector offers a promising avenue for ancillary services through Vehicle-to-Grid (V2G) applications. This is particularly critical for low-inertia systems, such as the UK grid, where the transition toward converter-based renewable generation necessitates very fast frequency response. Therefore, the viability of V2G for high-value frequency markets is constrained by strict latency requirements (e.g. one second)). Existing literature has predominantly focused on high-level economic aggregation models or communication network delays, largely neglecting the stochastic physical response dynamics of the EV On-Board Charger (OBC). This paper addresses this gap by developing a discrete-time Markov chain model that specifically characterizes the internal dynamics and response latency of OBC hardware. We integrate this model into a discrete-event simulation framework to evaluate end-to-end system latency, coupling stochastic OBC constraints with Over-the-Air (OTA) communication delays. We analyze the performance of fleets comprised of three common OBC ratings: 10 kW, 22 kW, and 43 kW. Contrary to the intuition that higher power ratings yield superior agility, our results demonstrate that high-capacity chargers may exhibit lower success rates in fast frequency markets due to insufficient ramp-rate-to-capacity ratios. Furthermore, we demonstrate that the frequency of mode switching events (switching between charging and discharging) is a dominant factor in performance degradation due to hardware hysteresis. These findings underscore that the efficacy of V2G applications requires precise EV-level control logic rather than relying solely on fleet-level optimization. Finally, the proposed models are evaluated against the PJM interconnection's composite score methodology. The results demonstrate high accuracy, suggesting the proposed framework can serve as a preliminary, EV-specific V2G assessment tool for market operators.