Memory access scheduling
Memory bandwidth
Dram
DOI:
10.1145/342001.339668
Publication Date:
2004-07-22T06:26:10Z
AUTHORS (5)
ABSTRACT
The bandwidth and latency of a memory system are strongly dependent on the manner in which accesses interact with “3-D” structure banks, rows, columns characteristic contemporary DRAM chips. There is nearly an order magnitude difference between successive references to different within row rows bank. This paper introduces access scheduling, technique that improves performance by reordering exploit locality 3-D structure. Conservative reordering, first ready reference sequence performed, 40% for traces from five media benchmarks. Aggressive operations scheduled optimize bandwidth, 93% same set applications. Memory scheduling particularly important processors where it enables processor make most efficient use scarce bandwidth.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (18)
CITATIONS (132)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....