Ultra-Low Latency with Java and Terabytes of Data
Could one create and consume a standard Java Stream in under 200 ns thereby paving the way for superfast data analytics? Just accessing 64-bit main memory takes about 100 ns and if the stream is backed by huge datasets, is it then even possible?
In this session, you will learn how to use the Speedment software that makes it possible to connect Java Streams directly to RAM allowing retrieval of data with ultra-low latency for a wide range of operations. This will work for Terabytes of data without impacting Garbage Collection. We will show how to leverage off-heap memory for storing serialized objects that are accessible via off-heap indexes and elaborate on the recent features in Java 9, 10 and 11 to learn how Dockerization, added Stream operation types, Local-Variable Type Inference and JVM-optimizations makes the solution even more efficient and easy to use.
The session will also cover more advanced features such as dynamically creating a joined Java stream from several tables and performing aggregations and reductions off-heap with no intermediary object creations, providing ultra-low latency also for these compound operations.