Infiniswap: Efficient Memory Disaggregation

Adrian Colyer covered this on the morning paper.

https://blog.acolyer.org/2017/05/05/efficient-memory-disaggr...

I once tried something like that for a very big montecarlo simulation of matrix convolutions for an online ads company.

Indeed, when you have no alternative to just have stuff in RAM as data access patterns are almost completely random for entire dataset, exporting swap over RDMA NIC will have an advantage even over a very fast local PCIE SSD.

I'm glad that somebody put time to make a proper solution to rationalize deployment of such setups