On Mon, Mar 16, 2009 at 2:34 PM, John David Duncan <[email protected]> wrote: >> >> JD, >> >> Am I correct in summarizing this by stating that MySQL has a fast >> version of this in code I can't review and problems with it are >> described in a problem report I cannot read? > > > Well, I think of it as a prototype. And I can only assume that the MySQL 6 > code is actually better and more suitable for public release than the > prototype that you can't see. The system calls used in the original Windows > (IOCP) and Solaris (port_associate(), etc.) prototypes are semantically > pretty different than libevent, or anything else available on Linux. > > I'm just saying a pool of threads *can* work acceptably.
I am sure it can, or someone I work with has wasted too many weeks backporting this code to 5.0.37. But there is a lot of work to get from here to there. Some of the changes are isolated to the pool-of-threads code. The 2X slowdown I get with pool-of-threads is likely caused by the global mutex used to schedule the threads. Architectural changes are also needed to keep a server from locking up when all threads in the pool are blocked on Innodb row locks (with a 50 second timeout in my case) or something else. A lot of MySQL users will be very unhappy if they try to use the 6.0 as is. -- Mark Callaghan [email protected] _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
