Deep learning (DL) has become a key component of modern software. Training DL models leads to huge carbon emissions. In data centers, it is important to reduce carbon emissions while completing DL training jobs early. In this article, we propose GreenFlow, a GPU cluster scheduler that reduces the average Job Completion Time (JCT) under a carbon emission budget. We first present performance models for DL training jobs to predict the throughput and energy consumption performance under different configurations. Based on the performance models and the carbon intensity of the grid, GreenFlow dynamically allocates GPUs, and adjusts the GPU-level and job-level configurations of DL training jobs. GreenFlow applies network packing and buddy allocation to job placement, thus avoiding extra carbon incurred by resource fragmentations. Evaluations on a real testbed show that when emitting the same amount of carbon, GreenFlow can improve the average JCT by up to 2.15×, compared to competitive baselines.

This paper proposes ElasticFlow, an elastic serverless training platform for distributed deep learning. ElasticFlow provides a serverless interface with two distinct features: (i) users specify only the deep neural network (DNN) model and hyperparameters for a job, but not the number of GPUs; (ii) users specify the deadline for a job, but not the amount of time to occupy GPUs. In contrast to existing server-centric platforms, ElasticFlow provides performance guarantees in terms of meeting deadlines while alleviating tedious, low-level, and manual resource management for deep learning developers. The characteristics of distributed training introduce two challenges. First, the training throughput scales non-linearly with the number of GPUs. Second, the scaling efficiency is affected by worker placement. To address these challenges, we propose Minimum Satisfactory Share to capture the resource usage of training jobs to meet deadlines, and ElasticFlow performs admission control based on it. We develop a greedy algorithm that dynamically allocates resources to admitted jobs based on diminishing returns. We apply buddy allocation to worker placement to eliminate the effect of topology. Evaluation results on a cluster of 128 GPUs show that ElasticFlow increases the number of jobs that can meet their deadlines by 1.46–7.65× compared to existing solutions.

Deep learning (DL) has become a key component of modern software. In the “ big model ” era, the rich features of DL-based software (i.e., DL software) substantially rely on powerful DL models, e.g., BERT, GPT-3, and the recently emerging GPT-4, which are trained on the powerful cloud with large datasets. Hence, training effective DL models has become a vital stage in the whole software lifecycle. When training deep learning models, especially those big models, developers need to parallelize and distribute the computation and memory resources amongst multiple devices (e.g., a cluster of GPUs) in the training process, which is known as distributed deep learning training , or distributed training for short. However, the unique challenges that developers encounter in distributed training process have not been studied in the software engineering community. Given the increasingly heavy dependence of current DL-based software on distributed training, this paper aims to fill in the knowledge gap and presents the first comprehensive study on developers’ issues in distributed training. To this end, we focus on popular DL frameworks that support distributed training (including TensorFlow, PyTorch, Keras, and Horovod) and analyze 1,131 real-world developers’ issues about using these frameworks reported on Stack Overflow and GitHub. We construct a fine-grained taxonomy consisting of 30 categories regarding the fault symptoms and summarize common fix patterns for different symptoms. We find that: (1) many distributed-specific faults and non-distributed-specific faults inherently share the same fault symptoms, making it challenging to debug; (2) most of the fault symptoms have frequent fix patterns; (3) about half of the faults are related to system-level configurations. Based on the results, we suggest actionable implications on research avenues that can potentially facilitate the distributed training to develop DL-based software, such as focusing on the frequent and common fix patterns when designing testing or debugging tools, developing efficient testing and debugging techniques for communication configuration along with the synthesis of network configuration analysis, designing new multi-device checkpoint-and-replay techniques to help reproduction, and designing serverless APIs for cloud platforms.

Mobile apps have been an integral part in our daily life. As these apps become more complex, it is critical to provide automated analysis techniques to ensure the correctness, security, and performance of these apps. A key component for these automated analysis techniques is to create a graphical user interface (GUI) model of an app, i.e., a window transition graph (WTG), that models windows and transitions among the windows. While existing work has provided both static and dynamic analysis to build the WTG for an app, the constructed WTG misses many transitions or contains many infeasible transitions due to the coverage issues of dynamic analysis and over-approximation of the static analysis. We propose ProMal, a "tribrid" analysis that synergistically combines static analysis, dynamic analysis, and machine learning to construct a precise WTG. Specifically, ProMal first applies static analysis to build a static WTG, and then applies dynamic analysis to verify the transitions in the static WTG. For the unverified transitions, ProMal further provides machine learning techniques that leverage runtime information (i.e., screenshots, UI layouts, and text information) to predict whether they are feasible transitions. Our evaluations on 40 real-world apps demonstrate the superiority of ProMal in building WTGs over static analysis, dynamic analysis, and machine learning techniques when they are applied separately.

Smartphone applications (a.k.a., apps) have become indispensable in our everyday life and work. In practice, accomplishing a task on smartphones may require the user to navigate among various apps. Unlike Web pages that are inherently interconnected through hyperlinks, apps are usually isolated building blocks, and the lack of direct links between apps has compromised the efficiency of task completion and user experience. In this article, we present the first in-depth empirical study of page-level access behaviors of smartphone users based on a comprehensive dataset collected through an extensive user study. We propose a model to distinguish informational pages and transitional pages , based on which we can extract page-level inter-app navigation. Surprisingly, the transitional pages account for quite substantial time cost and manual actions when navigating from the current informational page to the desirable informational page. We reveal that developing “ tunnels ” between “isolated” apps under specific usage scenarios has a huge potential to reduce the cost of navigation. Our analysis provides some practical implications on how to improve app-navigation experience from both the operating system’s perspective and the developer’s<?brk?> perspective.

Analyzing links among pages from different mobile apps is an important task of app analysis. Currently, most efforts of analyzing inter-app page links rely on static program analysis, which produces a lot of false positives, requiring significant manual effort to verify the links. To address the issue, in this paper, we propose LinkRadar, a data-driven approach to assisting the analysis of inter-app page links. Our key idea is to use dynamic program analysis to gather a set of actual inter-app page links, based on which we train a model to predict whether there exist links among pages from different apps to help verify the results of static program analysis. The challenge is that inter-app page links are hard to be triggered by dynamic program analysis, making it difficult to collect enough inter-app page links to train the model. Considering the similarity between intra-app page links and inter-app page links, we use transfer learning to deal with the data scarcity problem. Evaluation results show that LinkRadar is able to infer the inter-app page links with high accuracy.