Despite its popularity as a one-shot Neural Architecture Search (NAS) approach, the applicability of differentiable architecture search (DARTS) on complex vision tasks is still limited by the high computation and memory costs incurred by the over-parameterized supernet. We propose a new architecture search method called EasyNAS, whose memory and computational efficiency is achieved via our devised operator merging technique which shares and merges the weights of candidate convolution operations into a single convolution, and a dynamic channel refinement strategy. We also introduce a configurable search space-to-supernet conversion tool, leveraging the concept of atomic search components, to enable its application from classification to more complex vision tasks: detection and semantic segmentation. In classification, EasyNAS achieves state-of-the-art performance on the NAS-Bench-201 benchmark, attaining an impressive 76.2% accuracy on ImageNet. For detection, it achieves a mean average precision (mAP) of 40.1 with 120 frames per second (FPS) on MS-COCO test-dev. Additionally, we transfer the discovered architecture to the rotation detection task, where EasyNAS achieves a remarkable 77.05 mAP$_{50}$on the DOTA-v1.0 test set, using only 21.1 M parameters. In semantic segmentation, it achieves a competitive mean intersection over union (mIoU) of 72.6% at 173 FPS on Cityscape, after searching for only 0.7 GPU-day.

The innovation and evolution of file and storage systems have been influenced by workload analysis. Though cloud storage systems have been widely deployed and used, real-world and large-scale cloud storage workload studies are rare. Previous large-scale distributed storage systems can meet versatility, stability, and reliability requirements. Furthermore, modern cloud storage systems need to meet additional challenges, such as coping with surges in peak loads and rapid expansion of requests. These changes may lead to different characteristics.In this work, we propose DiTing data tracing system and collect workloads with over 242,000 billion requests from the Alibaba cloud. By comparing the normal days and the Single’s Day (the world’s largest online shopping festival), we analyze characteristics such as I/O scale, latency, locality, and load distribution. Our analysis reveals four key observations as follows. First, the virtual layer is the performance bottleneck of modern cloud storage systems during extreme peak periods. Second, the write operations dominate the data access because the application and operating system buffers absorb reads better than writes. Third, the workload is heavily skewed toward a small percentage of virtual cloud disks, with 20% of cloud disks accounting for 80% of I/O requests. Finally, data access shows poor temporal and spatial locality, and the I/O requests are mostly small-scaled. Based on these observations, we propose several suggestions for cloud storage systems, including separating I/O processing from the virtual layer to the proxy layer, deploying heavy and light workload applications on the same node, and adopting a write-friendly cloud disk design for write-skewed requests, etc. In summary, these workload characteristics and suggestions are useful for designing and implementing next-generation cloud storage systems.