Analysis of OLTP/OLAP Integrated Solutions: Apache Phoenix, Apache Trafodion, and Splice Machine

The paper starts by describing the need for systems that support both fast random reads/writes (OLTP) and large‑scale analytical queries (OLAP) in modern big‑data scenarios, introducing the concept of an Operational Data Store (ODS) that combines characteristics of data warehouses and transactional databases.

It then presents three prominent technologies that aim to fuse OLTP and OLAP capabilities: Apache Phoenix, Apache Trafodion, and Splice Machine. For each product, the article outlines its architecture, supported SQL standards, integration with HBase, query processing mechanisms, indexing options, transaction support, and performance characteristics.

Apache Phoenix is described as a SQL layer on top of HBase that translates SQL into native HBase scans, supports secondary indexes (global and local), statistics‑driven optimization, and ACID transactions via Apache Tephra.

Apache Trafodion is presented as a Hadoop/HBase‑based transactional SQL engine compatible with ANSI‑SQL, offering ODBC/JDBC drivers, a distributed connectivity service, cost‑based optimizer, and MPP execution engine.

Splice Machine is explained as a commercial product that combines HBase, Spark, and Derby, routing OLTP queries to HBase and OLAP queries to Spark, with its own optimizer, HBase coprocessor integration, and Spark resource management.

A comparative table (Table 1) summarizes the technical features of the three solutions. Following the comparison, the article details Baidu Waimai’s own ODS solution, which migrated from multiple MySQL instances to a Phoenix‑backed HBase store to achieve linear scalability, low‑latency writes, multi‑tenant isolation, and simplified query architecture. It also discusses implementation challenges such as the lack of bulk upsert support in Phoenix and the development of a custom Query Server to handle batch operations.

The conclusion reiterates the analysis of OLTP/OLAP integration, the evaluation of the three technologies, and the practical adoption of the Phoenix solution by Baidu Waimai, while briefly mentioning emerging alternatives like Apache Kudu.