As the adoption of LLMs becomes more widespread in software coding ecosystems, a pressing issue has emerged: does the generated code contain social bias and unfairness, such as those related to age, gender, and race? This issue concerns the integrity, fairness, and ethical foundation of software applications that depend on the code generated by these models but are underexplored in the literature. This article presents a novel bias testing framework that is specifically designed for code generation tasks. Based on this framework, we conduct an extensive empirical study on the biases in code generated by five widely studied LLMs (i.e., PALM-2-CodeChat-bison, Claude-instant-1, GPT-3.5-turbo, GPT-4-turbo, and GPT-4). Our findings reveal that biases are prevalent. For example, 13.47% to 49.10% of the codes generated by these LLMs have biased behaviors towards gender. Moreover, we study five bias mitigation prompt strategies that are commonly used in current code generation scenarios, i.e., zero-shot, one-shot, few-shot, and two Chain-of-Thought (CoT) prompts, with and without provided feedback-driven refinement. Our evaluation results illustrate that using direct prompt engineering strategies has limited effectiveness in mitigating bias, but our test execution feedback can help to reduce the ratio of code biases to a large extent (e.g., from 59.88% to 4.79% for GPT-4).
2 July 2021 in IEEE Transactions on Parallel and Distributed Systems
The increasing computational complexity of DNNs achieved unprecedented successes in various areas such as machine vision and natural language processing (NLP), e.g., the recent advanced Transformer has billions of parameters. However, as large-scale DNNs significantly exceed GPU's physical memory limit, they cannot be trained by conventional methods such as data parallelism. Pipeline parallelism that partitions a large DNN into small subnets and trains them on different GPUs is a plausible solution. Unfortunately, the layer partitioning and memory management in existing pipeline parallel systems are fixed during training, making them easily impeded by out-of-memory errors and the GPU under-utilization. These drawbacks amplify when performing neural architecture search (NAS) such as the evolved Transformer, where different network architectures of Transformer needed to be trained repeatedly. vPipe is the first system that transparently provides dynamic layer partitioning and memory management for pipeline parallelism. vPipe has two unique contributions, including (1) an online algorithm for searching a near-optimal layer partitioning and memory management plan, and (2) a live layer migration protocol for re-balancing the layer distribution across a training pipeline. vPipe improved the training throughput of two notable baselines (Pipedream and GPipe) by 61.4-463.4 percent and 24.8-291.3 percent on various large DNNs and training settings.
11 July 2022 in IEEE Transactions on Services Computing
Numerous blockchain systems with various consensus protocols have emerged to achieve high transaction rates (2$\sim$10K tps). However, their underlying P2P network primitives constrain further improvements due to two problems (i) high message redundancy and (ii) long broadcast convergence time. The first problem is caused by the excessive robustness of the dominant broadcast approach Gossip. All state-of-the-art blockchain systems only tolerate 20-50% node failure while Gossip can withstand up to 90%. The reason for (ii) is that existing broadcast topologies ignore geographical distances among nodes and incur paths with unnecessarily high latency. We present FRing, a geography-based P2P overlay network for fast and robust broadcast in blockchain systems. FRing has three main features: sufficient robustness, low message redundancy, and fast convergence. To reduce convergence time, FRing forms the network topology by considering geographical proximity. A novel broadcast algorithm based on FRing topology is proposed to lower message redundancy while maintaining sufficient robustness. One major challenge is to eliminate the risk of topology inference by traffic pattern analysis. FRing leverages Intel SGX to guarantee nodes’ behavior integrity and incorporates pattern obfuscation to prevent traffic pattern analysis. The evaluation shows that FRing improved the throughput of EOS by 220% and Hyperledger Fabric by 210%.
22 February 2023 in IEEE Transactions on Parallel and Distributed Systems
Training a large DNN (e.g., GPT3) efficiently on commodity clouds is challenging even with the latest 3D parallel training systems (e.g., Megatron v3.0). In particular, along the pipeline parallelism dimension, computational tasks that produce a whole DNN's gradients with multiple input batches should be concurrently activated; along the data parallelism dimension, a set of heavy-weight communications (for aggregating the accumulated outputs of computational tasks) is inevitably serialized after the pipelined tasks, undermining the training performance (e.g., in Megatron, data parallelism caused all GPUs idle for over 44% of the training time) over commodity cloud networks. To deserialize these communicational and computational tasks, we propose the AIAO scheduling (for 3D parallelism) which slices a DNN into multiple segments, so that the computational tasks processing the same DNN segment can be scheduled together, and the communicational tasks that synchronize this segment can be launched and overlapped (deserialized) with other segments’ computational tasks. We realized this idea in our Fold3D training system. Extensive evaluation shows Fold3D eliminated most of the all-GPU 44% idle time in Megatron (caused by data parallelism), leading to 25.2%–42.1% training throughput improvement compared to four notable baselines over various settings; Fold3D 's high performance scaled to many GPUs.
To provide applications with the ability to analyze fresh data and eliminate the time-consuming ETL workflow, hybrid transactional and analytical (HTAP) systems have been developed to serve online transaction processing and online analytical processing workloads in a single system. In recent years, HTAP systems have attracted considerable interest from both academia and industry. Several new architectures and technologies have been proposed. This paper provides a comprehensive overview of these HTAP systems. We review recently published papers and technical reports in this field and broadly classify existing HTAP systems into two categories based on their data formats: monolithic and hybrid HTAP. We further classify hybrid HTAP into four sub-categories based on their storage architecture: row-oriented, column-oriented, separated, and hybrid. Based on such a taxonomy, we outline each stream’s design challenges and performance issues (e.g., the contradictory format demand for monolithic HTAP). We then discuss potential solutions and their trade-offs by reviewing noteworthy research findings. Finally, we summarize emerging HTAP applications, benchmarks, future trends, and open problems.
While the Java Virtual Machine (JVM) plays a vital role in ensuring correct executions of Java applications, testing JVMs via generating and running class files on them can be rather challenging. The existing techniques, e.g., ClassFuzz and Classming, attempt to leverage the power of fuzzing and differential testing to cope with JVM intricacies by exposing discrepant execution results among different JVMs, i.e., inter-JVM discrepancies, for testing analytics. However, their adopted fuzzers are insufficiently guided since they include no well-designed seed and mutator scheduling mechanisms, leading to inefficient differential testing. To address such issues, in this paper, we propose SJFuzz, the first JVM fuzzing framework with seed and mutator scheduling mechanisms for automated JVM differential testing. Overall, SJFuzz aims to mutate class files via control flow mutators to facilitate the exposure of inter-JVM discrepancies. To this end, SJFuzz schedules seeds (class files) for mutations based on the discrepancy and diversity guidance. SJFuzz also schedules mutators for diversifying class file generation. To evaluate SJFuzz, we conduct an extensive study on multiple representative real-world JVMs, and the experimental results show that SJFuzz significantly outperforms the state-of-the-art mutation-based and generation-based JVM fuzzers in terms of the inter-JVM discrepancy exposure and the class file diversity. Moreover, SJFuzz successfully reported 46 potential JVM issues, and 20 of them have been confirmed as bugs and 16 have been fixed by the JVM developers.