On 08/19/2010 01:27 PM, Taras Glek wrote:
Hi,
I really appreciate that sqlite got this feature to reduce
fragmentation, but why not expose this as a pragma? In many cases it
is not feasible to pass the chunk size via a C API. For example with a
pragma I could do fragmentation testing via an sqlite shell, now this
option is out without modifying the shell source. I presume that a
similar problem exists for users of scripting languages until every
single scripting host provides support for this.
Perhaps this should even persist similar to how page_size persists
across connections.
Taras
To the patronizing "i don't believe this is a problem" and "even if it
is, Linux is superior to windows" concerns:
Attached my earlier email to the dev list showing that fragmented files
can be 3x slower to access than non-fragmented ones.
Trivia:
XFS on Linux gets a prize for being the most fragmentation-prone
filesystem in my survey. It is roughly 10-30x worse than everybody else.
Ext4 is ok, not great. Windows has a builtin autodefrag that on some
occasions ensures that our Firefox databases are not fragmented(ie it
beats the crap out of Linux contrary to the "classic" wisdom). Even
Apple's horrific filesystem occasionally results in less fragmented
files due to autodefrag.
For more details see
http://blog.mozilla.com/tglek/2010/07/22/file-fragmentation/
I already have a patch for Firefox that makes use of the new API, we
don't require a new pragma. It just seems like an important feature for
performance-tuning sqlite that isn't easy to get at for many users.
Taras
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
