On 22/09/14 10:48, Richard Hipp wrote:
> But if you have any new ideas on how we can further reduce the I/O, we'd love 
> to hear from you.

The single biggest problem for me is defaults.  SQLite supports memory
mapped i/o which has many advantages.  The stat4 analyze does a really good
job.  WAL reduces writing contention.  These are all off by default for
various good reasons.  But it also means that by default the majority of
developers are not getting the best performance SQLite already has to offer,
unless they happen to stumble on these.  Some like stat4 are especially
problematic since it requires recompilation to address.

It would be nice if by default things were best.  One weak suggestion is to
have a few profiles, and have them selectable by pragma.  The profile would
then go ahead and set various options.  Future versions of SQLite would also
then update the profiles.

For example:

  pragma profile='max_performance' -- turns on all above, ups caches etc
  pragma profile='min_memory'    -- tunes down everything related to memory

Various members of the SQLite consortium can then do things like:

  pragma profile='firefox'

sqlite-users mailing list

Reply via email to