Year: 2025
Type: conference-paper
Abstract: With computers processing an ever-increasing amount of data, expandingDRAM capacity is not always feasible due to economic andstructural issues. In this context, memory compression emergesas an alternative provided by the ZRAM module, which expandsthe DRAM’s data capacity by using a reserved area to store compressedpages directly into the main memory. Therefore, reducingsecondary storage accesses caused by page swaps and potentiallyreducing the program’s execution time. This paper evaluates theZRAM performance, comparing the speedup obtained when runningNAS Parallel Benchmarks under different disksize values forZRAM, in systems with 8 GB, 4 GB, and 1 GB of DRAM. Also, wecompared our baseline system with an older one, to study the effectof the systems improvement over the ZRAM usage impacts.Our results indicate that the impact of ZRAM on the application’sexecution time increases with the reduction of the available mainmemory space. We show no variations in speedup with 8 GB ofDRAM, and up to 10% improvement in scenarios with 4 GB ofmemory. In the 1 GB scenario, compression allowed certain benchmarksto achieve speed increases of up to 39%, even with a lowmemory footprint application, although most others did not benefit.We conclude that ZRAM has a more pronounced impact in memoryconstrainedenvironments, and can have a significant impact evenin applications with a low memory footprint. The code is availableat https://github.com/Guilherme-Eduardo/ZRAM.
Source: Anais do Computer on the Beach
Institution: Universidade Federal do Paraná
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4410770171