On Thu, Aug 14, 2008 at 10:24 AM, Jay Pipes <[EMAIL PROTECTED]> wrote: > Baron Schwartz wrote: >>> You can set UTF8 as default, but leave open the chance for people to >>> use other character sets. >> >> Given the current limitations of fixed-size buffers for sorting, >> group-by, and for internal temp tables (MEMORY engine) I would not set >> utf8 as the default anyway -- performance cost is too high. Although >> I understand the variable-sized MEMORY patches have been integrated? > > Understood, but if those current limitations were removed -- i.e. we > move to dynamically-sized buffers for sorting, group-by, etc, then will > this still be an issue? > >> Still, filesorts will suck if everything is just automatically utf8. > > Hmm, I'd have to actually see the performance degradation. If we remove > the logic around determining/setting charsets and kept (for now) only > the collation, I think some performance benefits can be realized. So, > only the collation logic/switching would be the performance issue.
How can you have a collation without a charset? If the fixed sized buffers are uh, fixed ;-) then I don't see as much of an issue. But some users will still see their tables get a lot bigger than they otherwise would be. _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
