From: Jeffrey J. K. <bac...@ko...> - 2009-11-17 03:40:37
|
Les Mikesell wrote at about 08:41:24 -0600 on Friday, November 13, 2009: > Peter Peltonen wrote: > > > >> Note that copying a whole pool (while preserving hardlinks) is still > >> tough and will only work up to a certain pool size/file count (apart > >> from copying a whole file system image). > > > > Do you have any more info on this (about the limits)? > > > > Why is it so, can't rsync handle big amounts of data including hardlinks? > > > > It is not just rsync - any file oriented technique will have the same problem. > The only way to match up the linked files is to build a table of filenames and > inode numbers as they are copied and look up the matching inode to find the > right name for the link. The process doesn't scale well. It may not be pretty because we don't have good tools for that process. But it scales as n log n which is not too bad (where n log n is the cost of sorting the table). And if my suggestion were followed of (optionally) appending the pool filename to the tail of each pool file then it would be O(n) and would be faster than even rsync without hard links. |