Most products at ByteDance, e.g., TikTok, Douyin, and Toutiao, naturally generate massive amounts of graph data. To efficiently store, query and update massive graph data is challenging for the broad range of products at ByteDance with various performance requirements. We categorize graph workloads at ByteDance into three types: online analytical, transaction, and serving processing, where each workload has its own characteristics. Existing graph databases have different performance bottlenecks in handling these workloads and none can efficiently handle the scale of graphs at ByteDance. We developed ByteGraph to process these graph workloads with high throughput, low latency and high scalability. There are several key designs in ByteGraph that make it efficient for processing our workloads, including edge-trees to store adjacency lists for high parallelism and low memory usage, adaptive optimizations on thread pools and indexes, and geographic replications to achieve fault tolerance and availability. ByteGraph has been in production use for several years and its performance has shown to be robust for processing a wide range of graph workloads at ByteDance.

Deep learning has been proposed for soft sensor modeling in process industries. However, conventional deep neural network (DNN) is a static network and thereby can not embrace evident dynamics in processes. Motivated by nonlinear autoregressive with exogenous input (NARX) model and neural nets based dynamic modeling, a dynamic network called NARX-DNN is put forward by further utilizing historical process samples and quality samples in a period of time. A modified hyper-parameter tuning method is proposed to choose optimal hyper-parameters of NARX-DNN with little manual intervention, which automatizes the training procedure and reduces computational cost. The quality prediction error of validation data is interpreted from different aspects, and the most appropriate delay of historical data can be determined automatically. The effectiveness of the proposed method is validated by case studies on a sulfur recovery unit and a debutanizer column. As training, validation and test data sets are selected by the original orders of data samples, the accurate prediction results of NARX-DNN demonstrate its ability in dealing with operation condition changes which are common in real processes.