Year: 2025
Type: conference-paper
Abstract: Memory requirements in the datacenter have increased dramatically, however current memory technologies are unable to keep up due to poor density scaling and pad-limited integration.Furthermore, such expensive memory resources are not efficiently utilized, in part, because of memory stranding issues.These factors have resulted in the emergence of disaggregated or far memory as a solution to increase capacity beyond a single node, and enable memory pooling to address underutilization.However, far memory comes with many challenges which may hinder its adoption.In this paper, we highlight these challenges ranging from high access latency, new failure domains, non-trivial data sharing mechanisms, and added infrastructure cost.We then enumerate a broad set of solutions to handle high access latency of far memory and introduce a taxonomy in terms of implementation complexity.Finally, we motivate an under explored solution-shipping code to data-using various function shipping techniques.
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4409714175