The rapid scaling of large language models (LLMs) has significantly increased GPU memory pressure, which is further aggravated by training optimization techniques such as virtual pipeline and recomputation that disrupt tensor lifespans and introduce considerable memory fragmentation. Such fragmentation stems from the use of online GPU memory allocators in popular deep learning frameworks like PyTorch, which disregard tensor lifespans. As a result, this inefficiency can waste as much as 43% of memory and trigger out-of-memory errors, undermining the effectiveness of optimization methods.

Democracy is crucial to a cryptocurrency ecosystem, as the diversity of miners (farms, personal computers, web clients, or even cloud functions) underlays the credibility of the cryptocurrency. Among miners, web clients used to be the vast majority, e.g., 50M+ as of March 2018. As time went on, however, cryptomining was gradually monopolized by mining farms with dedicated hardware (e.g., ASICs), and web clients scaled down to ∼0.1M. To suppress mining farms, certain cryptocurrencies (like Monero) adopted new mining algorithms such as RandomX whose execution relies on general-purpose hardware architectures. Unfortunately, this further impairs web-based cryptomining as web clients cannot provide the desired architecture support to these algorithms. This paper explores how to revive software democracy of efficient web-based crypto-mining, using a novel program transformation technique termed Vectra. Vectra employs just-in-time (JIT) transformations of mining programs for web architectures; it effectively identifies and merges isomorphic instructions upon execution. Vectra ensures correct transformations based on symbolic constraints of the instructions. Real-world deployments show that Vectra reduces WASM instructions by about 7× and achieves a 3× –16× speedup for web cryptomining in diverse execution environments like PCs, mobile phones, and serverless platforms, which translates to a high (69%–274%) return-on-investment (ROI) for common users.

Recent advances in mobile technologies (like WiFi 6 and 5G) do not seem to deliver the promised access bandwidth. To effectively characterize mobile access bandwidth in the wild, we work with a major commercial mobile bandwidth testing app to conduct a long-term (2020-2023) and large-scale (involving 4.76M users) measurement study in China, based on coarse-grained general statistics and fine-grained sampling diagnostics. Our study presents distinct facts as to WiFi, 5G, and 4G: in the past few years, the average WiFi download bandwidth exhibits a considerable rise (by 119.7% ), the average 5G download bandwidth constantly decreases (by a total of 20.2% ) despite the enormous infrastructure investments, while the average 4G download bandwidth first declines (by 22.1% ) and then increases (by 22.5% ). The situations of upload bandwidths are generally similar to those of download bandwidths, except that 5G upload bandwidths manifestN-shaped$(\nearrow \searrow \nearrow )$fluctuations. Our cross-layer and cross-technology analysis reveals a variety of impact factors as well as their complicated interplay as the root causes, such as the bottlenecks in underlying infrastructure (e.g., communication devices and wired Internet access), the traffic offloading from one access technology to another, the influence of the COVID-19 pandemic, and the side effects of aggressively migrating radio resources from 4G to 5G. With the longitudinal, holistic picture of today's mobile access bandwidth, we finally provide multifold practical implications on closing the technology gaps.

Modern mobile systems are featured by their increasing interactivity with users, which however is accompanied by a severe side effect---users constantly suffer from slow UI responsiveness (SUR). To date, the community have limited understandings of this issue for the challenges of comprehensively measuring SUR events on massive mobile devices. As a major Android phone vendor, in this paper we close the knowledge gap by conducting the first large-scale, long-term measurement study on SUR with 47M devices. Our study identifies the critical factors that lead to SUR from the perspectives of device, system, application, and app market. Most importantly, we note that the largest root cause lies in the wide existence of "hogging" apps, which persistently occupy an unreasonable amount of system resources by leveraging the optimistic design of Android process management. We have built on the insights to remodel Android process states by fully considering their time-sensitive transitions and the actual behaviors of processes, with remarkable real-world impact---the occurrences of SUR are reduced by 60%, together with 10.7% saving of battery consumption.

Emerging mobile apps such as UHD video and AR/VR access diverse high-throughput hardware devices, e.g., video codecs, cameras, and image processors. However, today's mobile emulators exhibit poor performance when emulating these devices. We pinpoint the major reason to be the discrepancy between the guest's and host's memory architectures for hardware devices, i.e., the mobile guest's centralized memory on a system-on-chip (SoC) versus the PC/server's separated memory modules on individual hardware. Such a discrepancy makes the shared virtual memory (SVM) architecture of mobile emulators highly inefficient.

Almost all Android users have ever experienced poor responsiveness, including the common frame dropping events—slow rendering (SR) and frozen frames (FF), as well as the uncommon Application Not Responding (ANR) and System Not Responding (SNR) that directly disrupt user experience. This work takes two complementary approaches,controlled benchmarkingandin-the-wild crowdsourcing, to comprehensively understand their prevalence, characteristics, and root causes, which turn out to be significantly different from common understandings and prior studies. We find that SR, FF, ANR, and SNR all occur prevalently on all the studied hardware models of Android phones, and better hardware does not seem to relieve ANR/SNR. Most surprisingly, they are oftentimes ascribed to defective software design that incurs substantial resource overuse—lightweight apps can experience severe SR/FF events due toredundant UI rendering, and the most ANR/SNR events stem from Android's aggressive implementation ofwrite amplification mitigation. In fact, the former can be effectively overcome by simplifying the apps' UI hierarchy, and we design a practical approach to address almost all ($>$99%) of the latter while only decreasing 3% of the data write speed with large-scale deployment. We have released our measurement code/data to the research community.

Mobile emulation, which creates full-fledged software mobile devices on a physical PC/server, is pivotal to the mobile ecosystem. Unfortunately, existing mobile emulators perform poorly on graphics-intensive apps in terms of efficiency and compatibility. To address this, we introduce graphics projection , a novel graphics virtualization mechanism that adds a small-size projection space inside the guest memory, which processes graphics operations involving control contexts and resource handles without host interactions. While enhancing performance, the decoupled and asynchronous guest/host control flows introduced by graphics projection can significantly complicate emulators’ reliability issue diagnosis when faced with a variety of uncommon or non-standard app behaviors in the wild, hindering practical deployment in production. To overcome this drawback, we develop an automatic reliability issue analysis pipeline that distills the critical code paths across the guest and host control flows by runtime quarantine and state introspection. The resulting new Android emulator, dubbed Trinity, exhibits an average of 97% native hardware performance and 99.3% reliable app support, in some cases outperforming other emulators by more than an order of magnitude.

Android system has been widely deployed in energy-constrained IoT devices for many practical applications, such as smart phone, smart home, healthcare, fitness, and beacons. However, Android users oftentimes suffer from app crashes, which directly disrupt user experience and could lead to data loss. Till now, the community have limited understanding of their prevalence, characteristics, and root causes. In this article, we make an in-depth study of the crash events regarding ten very popular apps of different genres, based on fine-grained system-level traces crowd-sourced from 93 million Android devices. We find that app crashes occur prevalently on the various hardware models studied, and better hardware does not seem to essentially relieve the problem. Most importantly, we unravel multi-fold root causes of app crashes, and pinpoint that the most crashes stem from the subtle yet crucial inconsistency between app developers’ supposed memory/process management model and Android’s actual implementations. We design practical approaches to addressing the inconsistency; after large-scale deployment, they reduce 40.4% of the app crashes with negligible system overhead. In addition, we summarize important lessons learned from this study, and have released our measurement code/data to the community.

Virtual devices based on device emulation have been widely used in lab research of mobile app testing for their efficiency and low cost. However, it remains controversial to use virtual devices for app testing in industry, given the inherent difficulties of high-fidelity emulation across diverse mobile systems and devices. Hence, mobile app companies still rely on physical device farms or services like AWS Device Farm.

Web performance optimization services, or web performance optimizers (WPOs), play a critical role in today’s web ecosystem by improving page load speed and saving network traffic. However, WPOs are known for introducing visual distortions that disrupt the users’ web experience. Unfortunately, visual distortions are hard to analyze, test, and debug, due to their subjective measure, dynamic content, and sophisticated WPO implementations.

Recent advances in mobile technologies such as 5G and WiFi 6E do not seem to deliver the promised mobile access bandwidth. To effectively characterize mobile access bandwidth in the wild, we work with a major commercial mobile bandwidth testing app to analyze mobile access bandwidths of 3.54M end users in China, based on fine-grained measurement and diagnostic information. Our analysis presents a surprising and frustrating fact---in the past two years, the average WiFi bandwidth remains largely unchanged, while the average 4G/5G bandwidth decreases remarkably. Our analysis further reveals the root causes---the bottlenecks in the underlying infrastructure (e.g., devices and wired Internet access) and side effects of aggressively migrating radio resources from 4G to 5G---with implications on closing the technology gaps. Additionally, our analysis provides insights on building ultra-fast, ultra-light bandwidth testing services (BTSes) at scale. Our new design dramatically reduces the test time of the commercial BTS from 10 seconds to 1 second on average, with a 15× reduction on the backend cost.

Androidoverlayenables one app to draw over other apps by creating an extraViewlayer atop the hostView, which nevertheless can be exploited by malicious apps (malware) to attack users. To combat this threat, prior countermeasures concentrate on restricting the capabilities of overlays at the OS level while sacrificing overlays’ usability; recently, the overlay mechanism has been substantially updated to prevent a variety of attacks, which however can still be evaded by considerable adversaries. To address these shortcomings, a more pragmatic approach is to enableearly detectionof overlay-based malware during the app market review process, so that all the capabilities of overlays can stay unchanged. For this purpose, in this paper we first conduct a large-scale comparative study of overlay characteristics in benign and malicious apps, and then implement the OverlayChecker system to automatically detect overlay-based malware for one of the world’s largest Android app stores. In particular, we have made systematic efforts in feature engineering, UI exploration, emulation architecture, and run-time environment, thus maintaining high detection accuracy (97 percent precision and 97 percent recall) and short per-app scan time ($\sim$1.7 minutes) with only two commodity servers, under an intensive workload of$\sim$10K newly submitted apps per day.

Delta synchronization (sync) is crucial to the network-level efficiency of cloud storage services, especially when handling large files with small increments. Practical delta sync techniques are, however, only available for PC clients and mobile apps, but not web browsers—the most pervasive and OS-independent access method. To bridge this gap, prior work concentrates on either reversing the delta sync protocol or utilizing the native client, all striving around the tradeoffs among efficiency, applicability, and usability and thus forming an “impossible triangle.” Recently, we note the advent of WebAssembly (WASM) , a portable binary instruction format that is efficient in both encoding size and load time. In principle, the unique advantages of WASM can make web-based applications enjoy near-native runtime speed without significant cloud-side or client-side changes. Thus, we implement a straightforward WASM-based delta sync solution, WASMrsync, finding its quasi-asynchronous working manner and conventional In-situ Separate Memory Allocation greatly increase sync time and memory usage. To address them, we strategically devise sync-async code decoupling and streaming compilation, together with Informed In-place File Construction. The resulting solution, WASMrsync+, achieves comparable sync time as the state-of-the-art (most efficient) solution with nearly only half of memory usage, letting the “impossible triangle” reach a reconciliation.

Carrying over 75% of the last-mile mobile Internet traffic, WiFi has inevitably become an enticing target for various security threats. In this work, we characterize a wide variety of real-world WiFi threats at an unprecedented scale, involving 19 million WiFi access points (APs) mostly located in China, by deploying a crowdsourced security checking system on 14 million mobile devices in the wild. Leveraging the collected data, we reveal the landscape of nationwide WiFi threats for the first time. We find that the prevalence, riskiness, and breakdown of WiFi threats deviate significantly from common understandings and prior studies. In particular, we detect attacks at around 4% of all WiFi APs, uncover that most WiFi attacks are driven by an underground economy, and provide strong evidence of web analytics platforms being the bottleneck of its monetization chain. Further, we provide insightful guidance for defending against WiFi attacks at scale, and some of our efforts have already yielded real-world impact---effectively disrupted the WiFi attack ecosystem.

With recent advances on cellular technologies (such as 5G) that push the boundary of cellular performance, cellular reliability has become a key concern of cellular technology adoption and deployment. However, this fundamental concern has never been addressed due to the challenges of measuring cellular reliability on mobile devices and the cost of conducting large-scale measurements. This paper closes the knowledge gap by presenting the first large-scale, in-depth study on cellular reliability with more than 70 million Android phones across 34 different hardware models. Our study identifies the critical factors that affect cellular reliability and clears up misleading intuitions indicated by common wisdom. In particular, our study pinpoints that software reliability defects are among the main root causes of cellular data connection failures. Our work provides actionable insights for improving cellular reliability at scale. More importantly, we have built on our insights to develop enhancements that effectively address cellular reliability issues with remarkable real-world impact---our optimizations on Android's cellular implementations have effectively reduced 40% cellular connection failures for 5G phones and 36% failure duration across all phones.

Despite being crucial to today's mobile ecosystem, app markets have meanwhile become a natural, convenient malware delivery channel as they actually “lend credibility” to malicious apps. In the past few years, machine learning (ML) techniques have been widely explored for automated, robust malware detection, but till now we have not seen an ML-based malware detection solution applied at market scales. To systematically understand the real-world challenges, we conduct a collaborative study with T-Market, a popular Android app market that offers us large-scale ground-truth data. Our study illustrates that the key to successfully developing such systems is multifold, including feature selection and encoding, feature engineering and exposure, app analysis speed and efficacy, developer and user engagement, as well as ML model evolution. Failure in any of the above aspects could lead to the “wooden barrel effect” of the whole system. This article presents our judicious design choices and first-hand deployment experiences in building a practical ML-powered malware detection system. It has been operational at T-Market, using a single commodity server to check ~12K apps every day, and has achieved an overall precision of 98.9 percent and recall of 98.1 percent with an average per-app scan time of 0.9 minutes.

App markets, being crucial and critical for today's mobile ecosystem, have also become a natural malware delivery channel since they actually "lend credibility" to malicious apps. In the past decade, machine learning (ML) techniques have been explored for automated, robust malware detection. Unfortunately, to date, we have yet to see an ML-based malware detection solution deployed at market scales. To better understand the real-world challenges, we conduct a collaborative study with a major Android app market (T-Market) offering us large-scale ground-truth data. Our study shows that the key to successfully developing such systems is manifold, including feature selection/engineering, app analysis speed, developer engagement, and model evolution. Failure in any of the above aspects would lead to the "wooden barrel effect" of the entire system. We discuss our careful design choices as well as our first-hand deployment experiences in building such an ML-powered malware detection system. We implement our design and examine its effectiveness in the T-Market for over one year, using a single commodity server to vet ~ 10K apps every day. The evaluation results show that this design achieves an overall precision of 98% and recall of 96% with an average per-app scan time of 1.3 minutes.

Almost every Android user has unsatisfying experiences regarding responsiveness, in particular Application Not Responding (ANR) and System Not Responding (SNR) that directly disrupt user experience. Unfortunately, the community have limited understanding of the prevalence, characteristics, and root causes of unresponsiveness. In this paper, we make an in-depth study of ANR and SNR at scale based on fine-grained system-level traces crowdsourced from 30,000 Android systems. We find that ANR and SNR occur prevalently on all the studied 15 hardware models, and better hardware does not seem to relieve the problem. Moreover, as Android evolves from version 7.0 to 9.0, there are fewer ANR events but more SNR events. Most importantly, we uncover multifold root causes of ANR and SNR and pinpoint the largest inefficiency which roots in Android's flawed implementation of Write Amplification Mitigation (WAM). We design a practical approach to eliminating this largest root cause; after large-scale deployment, it reduces almost all (>99%) ANR and SNR caused by WAM while only decreasing 3% of the data write speed. In addition, we document important lessons we have learned from this study, and have also released our measurement code/data to the research community.

Received Signal Strength (RSS) maps provide fundamental information for mobile users, aiding the development of conflict graph and improving communication quality to cope with the complex and unstable wireless channels. In this paper, we present CARM: a scheme that exploits crowd-sensing to construct outdoor RSS maps using smartphone measurements. An alternative yet impractical approach in literature is to appeal to professionals with customized devices. Our work distinguishes itself from previous studies by supporting off-the-shelf smartphone devices, and more importantly, by mitigating the error-prone nature and inaccuracies of these devices to build RSS maps through crowd-sensing. The main challenges are that, we need to calibrate error-prone smartphone measurements with “inaccurate” and “incomplete” data. To address these challenges, we build the measurement error model of smartphone based on the experimental observations and analyses. Moreover, we propose an iterative method based on Davidon-Fletcher-Powell (DFP) algorithm, to estimate the parameters for the error models of each smartphone and the signal propagation models of each AP simultaneously. The key intuition is that, the calibrated measurements based on the error model are constrained by the physics of the signal propagation model. Finally, a model-driven RSS map construction scheme is built upon these two models with these estimated parameters. The theoretical analyses prove the optimality and convergence of this iterative method. Also, the crowd-sensing experiments show that, CARM can achieve an accurate RSS map, decreasing the average error from 19.8 to 8.5 dBm.

We propose an accurate acoustic direction finding scheme, Swadloon, according to the arbitrary pattern of phone shaking in a rough horizontal plane. Swadloon leverages sensors of the smartphone without the requirement of any specialized devices. Our Swadloon design exploits a key observation: the relative displacement and velocity of the phone-shaking movement corresponds to the subtle phase and frequency shift of the Doppler effects experienced in the received acoustic signal by the phone. Swadloon tracks the displacement of smartphone relative to the acoustic direction with the resolution less than 1 millimeter. The direction is then obtained by combining the velocity from the displacement with the one from the inertial sensors. Major challenges in implementing Swadloon are to measure the displacement precisely and to estimate the shaking velocity accurately when the speed of phone-shaking is low and changes arbitrarily. We propose rigorous methods to address these challenges, and apply Swadloon to several case studies: Phone-to-Phone direction finding, indoor localization and tracking. Our extensive experiments show that the mean error of direction finding is around 2.1 degree within the range of 32 m. For indoor localization, the 90-percentile errors are under 0.92 m. For real-time tracking, the errors are within 0.4 m for walks of 51 m.

We propose an accurate acoustic direction finding scheme, Swadloon, according to the arbitrary pattern of phone shaking in rough horizontal plane. Swadloon tracks the displacement of smartphone relative to the acoustic direction with the resolution less than 1 millimeter. The direction is then obtained by combining the velocity from the displacement with the one from the inertial sensors. Major challenges in implementing Swadloon are to measure the displacement precisely and to estimate the shaking velocity accurately when the speed of phone-shaking is low and changes arbitrarily. We propose rigorous methods to address these challenges, and apply Swadloon to several case studies: Phone-to-Phone direction finding, indoor localization and tracking. Our extensive experiments show that the mean error of direction finding is around 2.1° within the range of 32 m. For indoor localization, the 90-percentile errors are under 0.92 m. For real-time tracking, the errors are within 0.4 m for walks of 51 m.