A modern data center is supported by an Internet of Assets (IoA), a specialized Internet of Things (IoT), where the sim-ready assets encompass physical properties and connected data streams for passive data collection and proactive simulation. A digital twin (DT) is a virtual replica of the IoA system with a proper level of abstraction, integrating asset-level dynamics models to simulate system-wide behavior, prototype designs, and conduct what-if analysis. However, creating high-fidelity DTs is hindered by the manual effort needed to encode complex geometric layouts and domain-specific design constraints. While large language models (LLMs) offer automation potential, existing methods struggle to generate geometrically plausible and functionally valid DT scenes due to limited domain integration. To bridge the gaps, we propose ChatDC, a conversational system that leverages LLMs to automate data center digital twin generation through a S egment- G enerate- O ptimize (SGO) workflow. ChatDC integrates domain knowledge via a dedicated code library named DCBuild and employs SGO to decompose user prompts, generate initial structures, and optimize layouts in compliance with data center design constraints. Evaluation shows that ChatDC outperforms other baselines with 98% success rate on scratch generation tasks and reduces the average makespan by 10×. Ablation study reveals that the SGO design increases the generation success rate by 65% at most. Furthermore, computational fluid dynamics simulations validate the physical plausibility, confirming the readiness of generated DTs for real-world analysis.

In mobile edge computing (MEC), one optimization strategy for mobile applications is to offload heavy computing tasks to cloud and edge servers. Constructing partitioning algorithms involves modeling individual methods through static code profilers, but exploiting dynamic user-driven execution patterns is also crucial. This paper introduces PartFlow, an interactive visualization system that supports comprehensive analysis of mobile application components and aids researchers in developing partitioning and offloading algorithms using real human behavioral data. PartFlow collects application component data remotely through binary instrumentation of mobile applications. Interactive diagrams are designed to evaluate component performance and illustrate transition patterns using the collected data. Additionally, PartFlow integrates a deep learning (DL)-based approach for multi-step forecasting of component states to improve accuracy and user experience in algorithm design. A case study and user feedback demonstrate PartFlow’s effectiveness in assisting researchers and engineers in creating offloading strategies.

Usability testing is vital for enhancing the user experience (UX) of interactive systems. However, analyzing test videos is complex and resource-intensive. Recent AI advancements have spurred exploration into human-AI collaboration for UX analysis, particularly through natural language. Unlike user-initiated dialogue, our study investigated the potential of proactive conversational assistants to aid UX evaluators through automatic suggestions at three distinct times: before, in sync with, and after potential usability problems. We conducted a hybrid Wizard-of-Oz study involving 24 UX evaluators, using ChatGPT to generate automatic problem suggestions and a human actor to respond to impromptu questions. While timing did not significantly impact analytic performance, suggestions appearing after potential problems were preferred, enhancing trust and efficiency. Participants found the automatic suggestions useful, but they collectively identified more than twice as many problems, underscoring the irreplaceable role of human expertise. Our findings also offer insights into future human-AI collaborative tools for UX evaluation.

The breakthrough of foundation models makes foundation model fine-tuning (FMF) workloads prevalent in modern GPU datacenters. However, existing schedulers tailored for model training do not consider the unique characteristics of FMs, making them inefficient in handling FMF workloads. To bridge the gap, we propose Ymir, a scheduler to improve the efficiency of FMF workloads in GPU datacenters. Ymir leverages the shared FM backbone architecture to expedite FMF workloads from two aspects: (1) Ymir investigates the task transferability among different FMF workloads and automatically merges FMF workloads with the same FM into one to improve the cluster-wide efficiency via transfer learning. (2) Ymir reuses the fine-tuning runtime of FMF workloads to reduce the significant context switch overhead. We conduct 32-GPU physical experiments and 240-GPU trace-driven simulations to validate the effectiveness of Ymir. Ymir can reduce the average job completion time by up to 4.3 × compared with existing state-of-the-art schedulers. It also promotes scheduling fairness by fully exploiting the task transferability. More supplementary materials can be found on our project website https://sites.google.com/view/ymir-project.

As mobile applications grow increasingly computation-intensive, the challenges arising from the limitations of mobile devices in terms of computing resources and battery life become more pronounced. Mobile Edge Computing (MEC) provides a promising avenue to address these challenges and enhance user experience. While existing studies have extensively explored resource allocation and task scheduling in MEC, most treat tasks as monolithic entities, overlooking the nuanced subtasks/components that often make up mobile applications. This paper endeavors to bridge the gap between the need for incentive mechanisms and the offloading of dependent computation tasks in MEC. Drawing inspiration from auction theory, we introduce a novel Multi-stage Iterative Combinatorial Double Auction (MICDA) mechanism, specifically tailored for dependent tasks in a cloud-edge-end cooperative computing scenario. Through theoretical analysis, the MICDA mechanisms demonstrate truthfulness, individual rationality, budget balance, and computational efficiency. Comprehensive experiment results further confirm its superior performance in improving application makespan and social welfare compared to other existing offloading strategies. This work validates the effective integration of dependency-aware computation offloading and auction mechanisms in overcoming economic and computational challenges in MEC systems, thereby paving the way for their potential application in broader real-world scenarios.

Mobile Edge Computing (MEC) enables computation offloading from resource-constrained mobile devices to edge servers in close vicinity, effectively promoting the user experience on emerging interactive multimedia applications such as virtual/augmented reality, mobile gaming, and mobile video editing. However, most contemporary MEC offloading research disregards the interdependencies between partitioned subtasks of application. Also, few studies focused on application topologies have neglected to design effective incentives to encourage edge servers to provide offloading services. In this paper, we propose a dependency-aware offloading algorithm based on a multi-round truthful combinatorial reverse auction (MTCRA) to address the social welfare maximization problem in the paradigm of MEC. Building on the topology of directed acyclic graphs (DAGs) modeled from applications, we discuss the complementarity and substitutability of subtasks in the context of combinatorial auction. Theoretical analysis shows that the presented auction mechanism achieves computing efficiency while maintaining desirable economic features like truthfulness, individual rationality, and budget balance. Simulation results demonstrate that the proposed algorithm achieves high social welfare regarding reduced execution time and good economic benefits for MEC servers.