tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: blocksizes


One place where that's currently not true is in the quote files,
there block counts are in blocks, and are 32 bits - making the counts
64 bits would double the size of the quota files (the entries really
want to remain a power of two bytes big), and halve cache efficiency
when reading them.   What's more, because we never try to squash two
files' data into the same disc sector (however big that is), we would
gain zero by counting bytes rather than blocks (every entry would simply
have N trailing zero bits)

While the premise of this last sentence is (and probably will be) correct,
the conclusion is not.  As NetBSD supports various filesystems, there may
be ones, that would profit from counting bytes instead of blocks.  In fact,
we already support one of those, of course currently without supporting
the features that would make use of this, namely the 9660 filesystem.
With this you can have files with multiple extents, where each extent's
last block contains some arbitrary number of bytes.  As an extreme
example, you could have a file with 3 bytes, where every byte is in a
separate block.  The idea ist to allow for append only files on some
multisession CD.  With the current block based buffer cache, support for
this would complicate the 9660 filesystem code extremely.

--                         Wolfgang Solfrank

Home | Main Index | Thread Index | Old Index