We are preparing to run a test restore of a client file system with nine million files using no-query restore. We are somewhat concerned about the amount of server memory needed to list all the files and sort them by location. How well or badly does this process behave if the server starts swapping? The traditional in-memory sort algorithms tend to have very large working set sizes, while some sort algorithms designed for virtual memory environments scan large blocks of data sequentially, resulting in much smaller working set sizes.
