Engineering of performance-sensitive applications requires a deep understanding of the interaction between hardware and software—particularly as modern hardware is increasingly complex. However, it is often challenging to couple findings from external profiling tools with knowledge from the application. This tutorial exemplifies how performance counters and sampling-based approaches can be leveraged within the application.

In-memory data processing is the state-of-the-art approach for large-scale data analytics. New developments in both memory technologies and application-specific accelerators introduce new opportunities and challenges to a computer system. To this end, Intel offers a new on-chip memory-operation accelerator called Data Streaming Accelerator (DSA) in modern Xeon Scalable processors, which enables higher data-transfer throughput by offloading memory operations from the CPU.

This tutorial introduces the DSA programming model and aims to enable participants to exploit the superior performance of the DSA over the CPU in terms of memory operations, e.g. to perform fast data relocation in a heterogeneous memory system.