Short-form video streaming is characterized by fast-paced, scrolling-driven user interactions. This poses unique challenges for its streaming algorithm design. To our knowledge, there is little understanding of how short video streaming algorithms perform in large-scale production platforms. To bridge this gap, this paper reports our two-year experience in evolving the prefetching algorithm, a critical algorithmic component for short video streaming, deployed in a leading global short video service. We adopt an iterative, production-driven approach, progressively evolving the design from simple heuristics to optimization-based and data-driven algorithms, with each iteration validated through large-scale A/B tests on hundreds of millions of users. Our evolution advances two core components: the prefetching logic and the viewing time estimation, including a lightweight on-device personalization mechanism. Through carefully balancing startup delay, mid-playback stalls, bandwidth usage, runtime overhead, and estimation accuracy, we achieve a 0.38% increase in user stay time - our key engagement metric - while simultaneously reducing bandwidth consumption by 14.7% throughout the evolution. We distill actionable insights from real-world deployment, highlighting the importance of startup latency, bandwidth efficiency, and low-overhead design for short video streaming at scale.

We present NIER, a video conferencing system that can adaptively maintain a low bitrate (e.g., 10–100 Kbps) with reasonable visual quality while being robust to packet losses. We use key-point-based deep image animation (DIA) as a key building block and address a series of networking and system challenges to make NIER practical. Our evaluations show that NIER significantly outperforms the baseline solutions.

The delivery of goods to numerous indoor stores poses significant safety risks, with heavy, high-stacked packages on delivery trolleys posing a potential hazard to passersby. This paper reports our experiences of developing and operating DeMo, a practical system for real-time monitoring of indoor delivery. DeMo employs sensors attached to trolleys, utilizing Inertial Measurement Unit (IMU) and Bluetooth Low Energy (BLE) readings to detect delivery violations, such as speeding and the use of non-designated delivery paths, and ensure accurate matching of each delivery to its intended destination store. Unlike typical indoor localization applications, DeMo addresses unique challenges, including sensor placement and the complex electromagnetic characteristics encountered in underground settings. Specifically, DeMo adapts the classical logarithmic radio signal model to facilitate fingerprint-free localization, significantly reducing deployment and maintenance costs. DeMo has been operating since May 2020, covering more than 200 shops with 74,537 deliveries (6193.2 km) across 12 subway stations in Hong Kong. DeMo's 4-year operation witnessed a significant violation rate drop, from 19% (May 2020) to 0.9% (Mar 2024).

People with Dementia (PwD) suffer from a high risk of getting lost due to their cognitive deterioration, leading to potential safety hazards and significant search efforts. In this paper, we propose DEmentia Caring System (DECS), an effective crowdsourcing platform to search for missing PwD. Specifically, PwD carry our customized Bluetooth Low Energy (BLE) tags that broadcast BLE packets, which are detected and then uploaded by mobile volunteers via their smartphones. To further enhance search efficiency, DECS deploys BLE gateways as its infrastructure and analyzes PwD's daily spatial-temporal mobility patterns. DECS has been deployed in Hong Kong since 2019, supporting 3,100+ PwD's families with over 45,000 app downloads by volunteers. More importantly, it has successfully served the search for 254 missing cases. This paper reports the unique lessons and experiences learned through our 4-year citywide deployment of DECS.

We organized an online fingerprint-based indoor localization competition in 2021. It attracted 1,170 teams worldwide. The teams were provided with a 60 GB dataset including WiFi, BLE, IMU, and geomagnetic field strength data collected from 204 buildings to build their localization algorithms, which were then evaluated against a separate test dataset. The competition received 28,009 submissions. The top team achieved an average accuracy of 1.50m. This paper reports the lessons we learned from analyzing the submissions, as well as our experiences in organizing the competition, through both qualitatively studying the teams' algorithms and quantitatively characterizing the competition results.

Network topology inference based on collected observations plays a fundamental role in many smart applications such as internet of vehicles. However, identifying the variation of network topology may be challenged since the network may be directed, and the topology inference problem is usually modelled as a non-convex problem. Moreover, large amount of data may be required to correctly identify the change of network topology. In this paper, the network topology is characterized by directed graphs, and the estimation of topology variation is modelled as a convex problem based on graph filtering. Since signals filtering over graphs can be considered to be a model-based problem, much less observations are required, and the topology estimation can then be performed with online behaviours. The proposed algorithms are validated on synthetic and real datasets, and both of them demonstrate good estimation performances.

We report our experiences of developing, deploying, and evaluating MLoc, a smartphone-based indoor localization system for malls. MLoc uses Bluetooth Low Energy RSSI and geomagnetic field strength as fingerprints. We develop efficient approaches for large-scale, outsourced training data collection. We also design robust online algorithms for localizing and tracking users' positions in complex malls. Since 2018, MLoc has been deployed in 7 cities in China, and used by more than 1 million customers. We conduct extensive evaluations at 35 malls in 7 cities, covering 152K m2 mall areas with a total walking distance of 215 km (1,100 km training data). MLoc yields a median location tracking error of 2.4m. We further characterize the behaviors of MLoc's customers (472K users visiting 12 malls), and demonstrate that MLoc is a promising marketing platform through a promotion event. The e-coupons delivered through MLoc yield an overall conversion rate of 22%. To facilitate future research on mobile sensing and indoor localization, we have released a large dataset (43 GB at the time when this paper was published) that contains IMU, BLE, GMF readings, and the localization ground truth collected by trained testers from 37 shopping malls.

More and more applications have deadline requirements for their data delivery such as 360° video, cloud VR gaming and autonomous driving. Those applications usually are band-width hungry. Fortunately, the data of those applications can be split into multiple blocks with different priorities making it possible to reduce the bandwidth consumption by prioritizing some blocks over others. However, the existing transport layer is too primitive to accomplish that. So those applications are forced to build their own customized and complex wheels. In this work, we propose Deadline-aware Transport Protocol (DTP) to provide deliver-before-deadline service. The application expresses the deadline and metadata of the data to DTP. Then DTP tries to meet the requirement by scheduling blocks. Compared to existing protocols, DTP provides meaningful service and reduces the burden of the application developer.

Cloud computing platforms enable applications to offer low-latency services to users by deploying data storage in multiple geo-distributed data centers. In this paper, through benchmark measurements on Amazon AWS and Microsoft Azure together with an analysis of a large-scale dataset collected from a major cloud CDN provider, we identify the high tail latency problem in cloud CDNs, which can substantially undermine the efficacy of cloud CDNs. One crucial idea to reduce the tail latency is to send requests in parallel to multiple clouds in cloud CDNs. However, since application providers often have a budget for using cloud services, deciding how many chunks to download from each cloud and when to download chunks in a cost-efficient manner still remain as open problems in our concerned scenario. To address the problem, we present TailCutter, a workload scheduling framework that aims at optimizing the tail latency while meeting cost constraints given by application providers. Specifically, we formulate the tail latency minimization (TLM) problem in cloud CDNs and design the receding horizon control based maximum tail minimization algorithm (RHC-based MTMA) to efficiently solve the TLM problem in practice. We implement TailCutter across multiple data centers of Amazon AWS and Microsoft Azure. Extensive evaluations using a large-scale real-world data trace (collected from a major ISP) illustrate that TailCutter can reduce up to 58.9% of the 100th-percentile user-perceived latency, as compared with alternative solutions under the cost constraint.