Evaluating the Impact of Memory Allocation in Persistent Memory Systems

• Grant number: 23/05032-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: May 01, 2023
• End date: April 30, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Alexandro José Baldassin
• Grantee: Otávio Scarparo Souza
• Host Institution: Instituto de Geociências e Ciências Exatas (IGCE). Universidade Estadual Paulista (UNESP). Campus de Rio Claro. Rio Claro , SP, Brazil
• Associated research grant: 18/15519-5 - Performance optimizations for multicore architectures, AP.JP2

Abstract

The recent introduction of byte-addressable non-volatile memory, or simply Persistent Memory (PM), has made it possible to develop data structures that are durable without using traditional file or database system interfaces. PM allows durability to be attained directly, without translating the data representation between the volatile and durable realms like traditional systems (i.e., serialization). One key element of any PM system is the memory allocator, responsible to manage the available unused memory. The allocator metadata itself has to be persistent, otherwise it might also encounter consistency issues. The research on memory allocation for PM system is a very hot topic and different approaches exist. The one used by the Intel PMDK (Persistent Memory Development Kit), for instance, provides an awkward interface in which a programmer has to provide an initialization function embedded in the call to the memory allocation. This occurs because the allocation and the assignment of the returned memory pointer must be atomic. Another alternative is to request memory (de)allocation only inside transactions. Regardless of the interface, the performance of the different memory allocators is still not well understood. In this scientific initiation project we seek to understand the performance impact of the main available memory allocators. To accomplish that, the execution times will be measured using standard memory-oriented workloads.