The hashes are located in NTDS.dit, although some software can inject into the LSASS process and extract them in-memory. In terms of which DC to use, this data should be replicated across each controller. You may see minor differences where a password has been changed on one and not yet copied the others (more on replication timings), but it is likely that the majority will be up-to-date.
quarks-pwdump expects you to use the Volume Shadow Copy method (utilising Volume Shadow Service - VSS) to retrieve NTDS.dit manually. The tool can then be used to parse hashes from this file. This in is contrast to dumping local hashes where the tool injects into the LSASS process. Using VSS (or registry backups for local dumping) can generally be safer and quieter than using an in-memory technique. These should be preferred where possible, which may be what Quarks are trying to promote.
The VSS method is detailed here (although this is to dump the registry hives, so needs to be adapted to NTDS.dit): http://bernardodamele.blogspot.co.uk/2011/12/dump-windows-password-hashes.html
A follow up post on NTDS.dit: http://bernardodamele.blogspot.co.uk/2011/12/dump-windows-password-hashes_16.html
Once you've retrieved the file, you can use quarks-pwdump to parse it with -dhd and -nt path/to/file/NTDS.dit (which is a hell of a lot easier than using NTDSXtract).
