AlltoAll communication in online Mixture-of-Experts inference inherently produces severe incast that can overwhelm transport, inflate switch queues, and raise tail latency. We present a lightweight, deployable online AlltoAll scheduler that aims to mitigate incast and return the system to a congestion-control-manageable regime, rather than pursuing global optimality. Our design integrates: (i) expert-load prediction; (ii) max-matching-based scheduling; and (iii) congestion-aware rescheduling. Preliminary experiments show our approach reduces AlltoAll completion time by up to 15.6%.

This paper reports a new stick-slip piezoelectric linear actuator with large load density, compact size, and high resolution. It is devised using a dual-arch bridge-type driving mechanism (DBDM) and an integrated elliptical preload mechanism(IEPM). The DBDM can generate a coupling motion on the driving point, which provides clamping and releasing actions during stick-slip motion. Also, the DBDM can magnify the clamping force with large amplification ratios, thus improving driving force and load-carrying capacity. The IEPM uses compact flexure hinges to adjust preload force, reducing structural dimension. Then, theoretical analysis and simulations are conducted, and a prototype is fabricated. A modified driving voltage with smooth turning points is designed to improve motion properties further. Experiments show that load density is 0.037 g/mm$^{\mathbf {3}}$under the driving voltage of 100 V. The actuator has an overall size of 33.6 mm$\times 20$mm$\times 8$mm and a displacement resolution of 22 nm.Note to Practitioners—Most research on linear actuators focused on motion displacement improvement, and more research needs to be done on improving load-carrying capacity. Therefore, this paper designs a force amplification linear actuator with a compact structure using the DBDM. In the experimental stage, the sudden change of the driving signal affects the output performance, and a modified sawtooth voltage is used to improve the output speed and load-carrying capacity. The experimental results show that under the modified sawtooth wave voltage, the maximum speed of the slider is 1.4 mm/s, and the speed is increased by 17% compared with the traditional wave. When a 200 g load is applied to the slider, the linear actuator operates at a stable speed of$1.2~\mu $m/s, 33% faster than conventional sawtooth waves. As a result, the designed actuator has superior performance at higher loads.

Pervasive applications are involving more and more autonomous computing and communicating devices, augmented with the abilities of sensing and controlling the logical/physical environment. To enable context-awareness for such applications, we are challenged by the intrinsic asynchrony of the computing environment. Predicate detection is a well studied technique dedicated to detecting global predicates over asynchronous computations and can be employed to achieve context-awareness of the asynchronous environment. However, there is no methodological framework which guides us to systematically apply the abstract predicate detection theory to the development of concrete context-aware applications. To this end, we present the Predicate Detection-based ContextAwareness (PD-CA) framework. PD-CA maps the concepts of context-awareness to concepts of predicate detection. PD-CA also presents a design process of providing middleware support for context-aware applications. Under the guidance of the PD-CA framework, we design and implement the Middleware Infrastructure for Predicate detection in Asynchronous environments (MIPA). We also propose the programming toolkit to facilitate the development of context-aware applications based on MIPA, and demonstrate the use of the toolkit by a case study of a chemical plant safety management application. Experimental evaluations show the performance of MIPA in enabling context-awareness despite of the asynchrony.

The event stream model of computation has a wide range of applications, e.g, computer system monitoring, physical environment sensing/surveillance, and stock trade monitoring. Sliding windows are widely used to facilitate effective event stream processing. However, it is greatly challenged when the event sources are distributed and asynchronous. One important technique to cope with the asynchrony is to utilize that the meaningful snapshots of an asynchronous computation form a distributive lattice. It thus becomes the central challenge whether this lattice structure still preserves and how to maintain it at runtime, when we restrict our attention to events within sliding windows. To address this challenge, we first prove that the snapshots of the asynchronous event streams within the sliding windows form a convex distributive lattice (denoted by Lat-Win). This enables us to easily integrate existing predicate specification and detection techniques, to express and monitor properties of our concern over asynchronous event streams. Then we propose an algorithm to maintain Lat-Win at runtime. The proposed scheme is evaluated in a context-aware smart office scenario, where activities of the user can be recognized by monitoring multiple streams of sensed events. The Lat-Win algorithm is implemented on the open-source context-aware middleware we developed. The evaluation results first show the advantage of adopting sliding windows over asynchronous event streams. Then they show the performance of detecting specified predicates within Lat-Win, with dynamic changes in the computing environment.

Formal specification and runtime detection of contextual properties is one of the primary approaches to enabling context awareness in pervasive computing environments. Due to the intrinsic dynamism of the pervasive computing environment, dynamic properties, which delineate concerns of context-aware applications on the temporal evolution of the environment state, are of great importance. However, detection of dynamic properties is challenging, mainly due to the intrinsic asynchrony among computing entities in the pervasive computing environment. Moreover, the detection must be conducted at runtime in pervasive computing scenarios, which makes existing schemes do not work. To address these challenges, we propose the property detection for asynchronous context (PDAC) framework, which consists of three essential parts: 1) Logical time is employed to model the temporal evolution of environment state as a lattice. The active surface of the lattice is introduced as the key notion to model the runtime evolution of the environment state; 2) Specification of dynamic properties is viewed as a formal language defined over the trace of environment state evolution; and 3) The SurfMaint algorithm is proposed to achieve runtime maintenance of the active surface of the lattice, which further enables runtime detection of dynamic properties. A case study is conducted to demonstrate how the PDAC framework enables context awareness in asynchronous pervasive computing scenarios. The SurfMaint algorithm is implemented and evaluated over MIPA--the open-source context-aware middleware we developed. Performance measurements show the accuracy and cost-effectiveness of SurfMaint, even when faced with dynamic changes in the asynchronous pervasive computing environment.

Efficient detection of RFID-tagged physical objects is one of the key enabling technologies to build pervasive Internet-of-Things applications. However, the detection of tagged-objects is faced with the critical challenge of tag dynamics, which mainly arises from the movement of tagged physical objects. To capture tag dynamics, the application needs to detect the presence/absence of tags in an accurate, timely and cost-effective way. To address these challenges, we propose the Prediction of Tag Dynamics (PTD) algorithm. PTD achieves runtime detection of tagged-objects by i) streaming of the temporally-correlated tag readings obtained from persistent tracking of the tagged-object, and ii) runtime prediction of tag dynamics based on the streaming of tag readings. The performance of PTD is investigated based on real implementation and experimental evaluation, where PTD processes tag readings gathered with high fidelity from persistent tracking of real activities of tagged objects. The evaluation results demonstrate the accuracy, timeliness and cost-effectiveness of PTD.

Runtime detection of contextual properties is one of the primary approaches to enabling context-awareness in pervasive computing scenarios. Among various properties the applications may specify, the concurrency property, i.e., property delineating concurrency among contextual activities, is of great importance. It is because the concurrency property is one of the most frequently specified properties by context-aware applications. Moreover, the concurrency property serves as the basis for specification of many other properties. Existing schemes implicitly assume that context collecting devices share the same notion of time. Thus, the concurrency property can be easily detected. However, this assumption does not necessarily hold in pervasive computing environments, which are characterized by the asynchronous coordination among heterogeneous computing entities. To cope with this challenge, we identify and address three essential issues. First, we introduce logical time to model behavior of the asynchronous pervasive computing environment. Second, we propose the logic for specification of the concurrency property. Third, we propose the Concurrent contextual Activity Detection in Asynchronous environments (CADA) algorithm, which achieves runtime detection of the concurrency property. Performance analysis and experimental evaluation show that CADA effectively detects the concurrency property in asynchronous pervasive computing scenarios.

This paper studies the problem of categorical data clustering, especially for transactional data characterized by high dimensionality and large volume. Starting from a heuristic method of increasing the height-to-width ratio of the cluster histogram, we develop a novel algorithm -- CLOPE, which is very fast and scalable, while being quite effective. We demonstrate the performance of our algorithm on two real world datasets, and compare CLOPE with the state-of-art algorithms.