Publications

2026

1. VeCT: Secure and Efficient Constant-Time Code Rewriting with Vector Extensions

   Qisheng Jiang and Danfeng Zhang

   In 35th USENIX Security Symposium (USENIX Security 26), 2026

   Abstract PDF Code Website

   Timing channels allow attackers to extract secrets by analyzing the execution time of a victim program. Constant-time (CT) disciplines enforce security against timing attacks via data-flow/control-flow linearization (DFL/CFL). However, the rewritten constant-time code typically considerably increases the memory footprint of the original code, causing significant overhead. We present VeCT, a compiler-based code rewriter that leverages vector extensions to retain constant-time guarantees while improving performance. We first apply rigorous statistical tests to derive practical “safe-use” rules for AVX-512 instructions whose implementation details are proprietary; this analysis also reveals a previously unknown vulnerability in a state-of-the-art constant-time rewriter. Guided by these rules, VeCT introduces a novel strategy that eliminates unnecessary data loads in rewritten code, and enables vectorization to further improve efficiency. We implement VeCT based on LLVM to automatically transform code into AVX-512-based constant-time equivalents. On real-world applications like AES and Blowfish, VeCT reduces the overhead of transformed code by up to 98.9% compared to the state-of-the-art, while preserving constant-time behavior.

2024

1. Sync+ Sync: A Covert Channel Built on fsync with Storage

   Qisheng Jiang and Chundong Wang

   In 33rd USENIX Security Symposium (USENIX Security 24), 2024

   Abstract Code Website

   Scientists have built a variety of covert channels for secretive information transmission with CPU cache and main memory. In this paper, we turn to a lower level in the memory hierarchy, i.e., persistent storage. Most programs store intermediate or eventual results in the form of files and some of them call fsync to synchronously persist a file with storage device for orderly persistence. Our quantitative study shows that one program would undergo significantly longer response time for fsync call if the other program is concurrently calling fsync, although they do not share any data. We further find that, concurrent fsync calls contend at multiple levels of storage stack due to sharing software structures (e.g., Ext4’s journal) and hardware resources (e.g., disk’s I/O dispatch queue). We accordingly build a covert channel named Sync+Sync. Sync+Sync delivers a transmission bandwidth of 20,000 bits per second at an error rate of about 0.40% with an ordinary solid-state drive. Sync+Sync can be conducted in cross-disk partition, cross-file system, cross-container, cross-virtual machine, and even cross-disk drive fashions, without sharing data between programs. Next, we launch side-channel attacks with Sync+Sync and manage to precisely detect operations of a victim database (e.g., insert/update and B-Tree node split). We also leverage Sync+Sync to distinguish applications and websites with high accuracy by detecting and analyzing their fsync frequencies and flushed data volumes. These attacks are useful to support further fine-grained information leakage.

2. GNNDrive: Reducing Memory Contention and I/O Congestion for Disk-based GNN Training

   Qisheng Jiang, Lei Jia, and Chundong Wang

   In Proceedings of the 53rd International Conference on Parallel Processing, 2024

   Abstract Code Website

   Graph neural networks (GNNs) gain wide popularity. Large graphs with high-dimensional features become common and training GNNs on them is non-trivial on an ordinary machine. Given a gigantic graph, even sample-based GNN training cannot work efficiently, since it is difficult to keep the graph’s entire data in memory during the training process. Leveraging a solid-state drive (SSD) or other storage devices to extend the memory space has been studied in training GNNs. Memory and I/Os are hence critical for effectual disk-based training. We find that state-of-the-art (SoTA) disk-based GNN training systems severely suffer from issues like the memory contention between a graph’s topological and feature data, and severe I/O congestion upon loading data from SSD for training. We accordingly develop GNNDrive. GNNDrive 1) minimizes the memory footprint with holistic buffer management across sampling and extracting, and 2) avoids I/O congestion through a strategy of asynchronous feature extraction. It also avoids costly data preparation on the critical path and makes the most of software and hardware resources. Experiments show that GNNDrive achieves superior performance. For example, when training with the Papers100M dataset and GraphSAGE model, GNNDrive is faster than SoTA PyG+, Ginex, and MariusGNN by 16.9x, 2.6x, and 2.7x, respectively.

3. Hercules: Enabling Atomic Durability for Persistent Memory with Transient Persistence Domain

   Chongnan Ye, Meng Chen, Qisheng Jiang, and 1 more author

   ACM Transactions on Embedded Computing Systems, 2024

   Website
4. Caiti: I/O transit caching for persistent memory-based block device

   Qing Xu, Qisheng Jiang, and Chundong Wang

   Journal of Systems Architecture, 2024

   Website

2023

1. Atomic but Lazy Updating with Memory-mapped Files for Persistent Memory

   Qisheng Jiang, Lei Jia, and Chundong Wang

   In 2023 Design, Automation & Test in Europe Conference & Exhibition (DATE), 2023

   Abstract Slides Website

   Applications memory-map file data stored in the persistent memory and expect both high performance and failure atomicity. State-of-the-art NOVA and Libnvmmio guarantee failure atomicity but yield inferior performance. They enforce data staying fresh and intact at the mapped addresses by continually updating the data there, thereby incurring severe write amplifications. They also lack the adaptability to dynamic workloads and entail housekeeping overheads with complex designs. We hence propose Acumen with a group of reflection pages managed for a mapped file. Using a simplistic bitmap to track fine-grained data slices, Acumen makes a reflection page and a mapped file page pair to alternately carry updates to achieve failure atomicity. Only on receiving a read request will it deploy valid data from reflection pages into target mapped file pages. The cost of deployment is amortized over subsequent read requests. Experiments show that Acumen significantly outperforms NOVA and Libnvmmio with consistently higher performance in serving a variety of workloads.

2. Exploring architectural implications to boost performance for in-nvm b+-tree

   Yanpeng Hu, Qisheng Jiang, and Chundong Wang

   In Proceedings of the 28th Asia and South Pacific Design Automation Conference, 2023

   Website

2022

1. Early Forecasting of the Impact of Traffic Accidents Using a Single Shot Observation

   Guangyu Meng, Qisheng Jiang, Kaiqun Fu, and 3 more authors

   In Proceedings of the 2022 SIAM International Conference on Data Mining (SDM), 2022

   Website