Remote Memory is neat in that application can use memory provided by remote servers just as if they are presented as local DRAM memory. It has regained focus thanks to emerging ultra-low latency RDMA (Remote Direct Memory Access) network equipment. Hardware, software or a mix of both can be used to construct Remote Memory solutions, each with respectively advantages and costs. The software-based approach is attractive in that it may require no extra hardware, with no application rewrite or programming change. In this talk, we will introduce the concept, use cases and experiences of software-based remote memory. Then we will discuss some future optimization directions.
Collaborative note: https://hackmd.io/@coscup/rkYxCGl4S