Case Study: Netease boosts ad hoc SQL performance with Alluxio and Presto

How Netease Games leveraged Alluxio to boost ad hoc SQL on HDFS

Netease Games, a leading online game company, needed a faster way to support ad hoc SQL analytics on massive HDFS-based data sets. Its existing Hadoop stack, including Hive on MapReduce, Hive on Tez, SparkSQL, and HBase-based approaches, could not consistently meet the desired 2–15 second response time for interactive queries, especially under peak load and with unstable HDFS performance.

Netease Games implemented a Presto + Alluxio solution from Alluxio, deploying Presto and Alluxio on a separate satellite cluster with tiered caching and synchronized Hive metadata. This allowed queries to run directly on Hive data without ETL, improved stability by caching hot data in Alluxio, and provided progress visibility for users. The result was significantly better and more stable query performance, with query latency reduced to around 10 seconds or less on average and much lower variance.

Netease