The Raft consensus protocol naturally fits time series databases, owing to the resemblance between its continuous log and the time series data. While the serialization of appending entries reduces the state space for ease design and implementation, it blocks the subsequent requests and thus limits the parallelism and throughput of Raft. Intuitively, once an entry arrives the follower, we may notice the leader and the client to unblock the subsequent as early, rather than waiting for its appending and committing. In this way, more requests can be processed in parallel, and thus the throughput increases, essential for IoT applications often with vast sensors and fast data ingestion. Of course, with higher parallelism, the risk of persistence for in-processing entries increases. It is a worthwhile trade-off in the IoT scenario since tiny data loss during leader failure is more acceptable than shutting out most data due to a low throughput. Our Non-Blocking Raft (NB-Raft) is implemented as the consensus protocol of Apache IoTDB, a commodity time series database management system, supporting various applications in Alibaba Cloud. Extensive evaluation shows that the throughput is improved by about 30% using our NB-Raft compared to the original Raft, a considerable amount of further data saved.

The server workloads of large-scale online video games are elastic and on-demand. The workload can range from tens to thousands of server instances in short periods. In fact, a cloud-based video game ecosystem can reach a workload of millions of players every week. Given such a large scale, even a small portion of over-provisioning leads to a significant amount of resource idling and a high cost of waste. It is essential to define an effective forecasting model on the game session workloads given a time span. The effectiveness shall be measured by metrics representing Service Level Objectives (SLOs). In this work, we analyze time series forecasting models using ARIMA, Prophet, and LSTM to predict the number of virtual machines in need of cloud resource monitoring data. In addition, we define service-level metrics for measuring effectiveness based on factors of over/under provision and ratio of resource waste. We analyze models with 16 fleets with an average of 2754 game servers over a four-month-long period of time in the production environment. We observe that our LSTM model is the most accurate in forecasting the demand of virtual machines in terms of RMSE and MAE. Further analysis using metrics of SLOs, we observe that the LSTM model leads to more cases of under-provisioning than ARIMA and Prophet do. The LSTM model forecasts the demand of virtual machines with less over-provision ratio than ARIMA and Prophet do for 14 out of 16 fleets. Using the LSTM model, we further evaluate the forecasting effect across different time spans of a single fleet and across multiple fleets within the same time span.