On Fri, 18 May 2018, Jason Gunthorpe wrote: > Ummm, RDMA has done essentially that since 2005, since when did it > become wrong? Do you have some references? Is there some alternative? It was wrong from the start. It became much more evident with widespread use of RDMA. The inability to scale processor performance at this point but the huge increase in network bandwidth available forces users into RDMA solution. Thus they will try to do RDMA to file backed mappings where in the past we only used anonymous memory for RDMA.
