[parted-devel] Failure with O_DIRECT

Jim Meyering jim at meyering.net
Wed Jul 4 14:06:06 UTC 2007


"Leslie P. Polzer" <leslie.polzer at gmx.net> wrote:
> phys_sector_size is the drive's physical (internal) sector size, a thing
> reported since ATAPI8 IIRC (see http://bigsector.org/ for information on this
> topic). Alignment need not be aligned to this, but the should, to avoid
> severe speed penalties.  Parted seems to be the only partitioning tool
> to support this at the moment.

Thanks for the explanation.

> sector_size is the drive's logical sector size.

This must be a multiple of phys_sector_size, right?

> Both of them have nothing to do with the file system's block size;

Yes, I see that's true when we're talking about a real device, but...

> the three constraints are imposed separately on alignment calculations.

...when using a regular file (init_file), sector_size is set
to PED_SECTOR_SIZE_DEFAULT unconditionally, regardless of the
underlying file system block size.  But here, it matters for
O_DIRECT on losing (e.g. 2.4.x) systems.



More information about the parted-devel mailing list