The paper 'Performance Evaluation and Analysis for LSM based Persistent Key-Value Store in Streaming Application' by Park Gyu-ri, graduate student in the Department of Computer Science & Engineering (supervisor: Park Sung-yong), received the Outstanding Presentation Paper Award at the Korea Computer Congress 2023 (KCC 2023), held over four days from 17 to 20 June.

▲ From left: Park Gyu-ri (master's, 3rd semester), Jung Yeon-woo (doctoral, 7th semester), Kim Ji-seop (master's, 1st semester), Park Sung-yong (supervisor)

▲ Certificate of the Outstanding Presentation Paper Award, Korea Computer Congress 2023 (KCC 2023)

As demands have risen for processing large volumes of streaming data and stateful queries, embedded persistent key-value stores based on log-structured merge (LSM) trees, which maintain large amounts of state, have emerged. LSM operations for state management improve stateful query performance but affect overall performance through disk I/O. This paper analyses the performance patterns and factors arising from LSM operations when an LSM-based persistent key-value store is used as the state engine in the Apache Spark framework. The experiments compare performance patterns with the compaction threshold — an LSM operation — set to 4, 16 and 64. Overall, the study confirms that with short compaction intervals compaction work is the main factor in performance degradation, while with long compaction intervals read operations are the main factor; it also confirms that as LSM operations are delayed, incoming data accumulates and batch sizes grow, degrading query performance.

The Korea Computer Congress 2023 (KCC 2023) is an information science and technology conference at which the latest research results across all fields of information technology — artificial intelligence, computer systems, databases and more — are presented.