=================================================================================

Tungsten is part of Spark's effort to improve the efficiency of memory and CPU for Spark applications, enabling Spark to handle much larger data sets and improving the performance of processing-intensive applications. Key features of Tungsten include:

• Memory Management and Binary Processing: Bypasses the JVM (Java Virtual Machine) object model and garbage collection by using off-heap memory management for more predictable performance and by optimizing the representation of data in memory. Data is processed in a serialized binary format to reduce the overhead of JVM memory structure. This means it stores data in a format that is more compact and easier for Spark to process, which minimizes the overhead of garbage collection.
• Cache-aware computation: Optimizes algorithms and data structures to effectively utilize CPU caches.
• Whole-stage code generation: A physical query optimization phase that fuses multiple operations together into a single function, eliminating virtual function calls and reducing JVM overhead. This is done by dynamically generating compact bytecode, for the entire query stage, that runs faster than the standard Java bytecode execution.

===========================================