There is still one possibility how to make commit-merge faster. Example, instead of dividing virtual disks by snapshots among different files, as with vmware, there can be only one file. Metadata knows where are blocks of one snapshot data and where are others. Now, instead of copying while commit, it just changes pointers to those blocks. This also means data becomes fragmented. In one day it still needs to defragment it. Maybe in background. So, there is actually no big difference. Altough separating snapshots by different files looks safer, because if something wrong goes with metadata or pointers, all stuff is out.
↧