Transformers have been widely used in various computer vision applications. Compared to traditional convolutional neural networks (CNNs), transformer's inference includes plenty of non‐linear operations, such as softmax and Gaussian error linear units (GELU). As the scale of transformers grows, an efficient hardware implementation of these operations is significant. However, the current works of computer vision neural network accelerators focus on CNN and less attention is paid to transformer. In addition, most current FPGA‐based softmax or GELU accelerators are not designed for vision transformer (ViT). To solve this problem, this work proposes a high speed reconfigurable accelerator. The architecture can support both softmax and GELU functions in ViT by reconfiguring the data path. This architecture on Xilinx XCVU37P is implemented through mathematical transformation and hardware optimization design, and achieve the performance of 102.4 Giga bits per second (Gbps) at 200 MHz. Experimental results show that the architecture achieves a very small accuracy loss in the ViT's inference by using fixed‐point 16‐bit quantization. Compared with existing accelerators, the design has greater throughput and area efficiency.

With the growing amount of information and data, object-oriented storage systems have been widely used in many applications, including the Google File System, Amazon S3, Hadoop Distributed File System, and Ceph, in which load balancing of metadata plays an important role in improving the input/output performance of the entire system. Unbalanced load on the metadata server leads to a serious bottleneck problem for system performance. However, most existing metadata load balancing strategies, which are based on subtree segmentation or hashing, lack good dynamics and adaptability. In this study, we propose a metadata dynamic load balancing (MDLB) mechanism based on reinforcement learning (RL). We learn that the Q_learning algorithm and our RL-based strategy consist of three modules, i.e., the policy selection network, load balancing network, and parameter update network. Experimental results show that the proposed MDLB algorithm can adjust the load dynamically according to the performance of the metadata servers, and that it has good adaptability in the case of sudden change of data volume.

No abstract

No abstract