We're running mostly with InnoDB tables, about 5% updates/inserts/deletes, the rest selects, on Windows NT.
In setting table_cache to 256 from the default 64, we hoped to improve performance a little, by not having to continually close/open tables. Then we noticed that the opened table count dropped to zero and began to climb again every 30 minutes - a consequence, through later reading of the Fine Manual, of the flush_time setting of 30 minutes (1800 sec), which seems to be recommended for W9x and Me only. The docs state that this action 'closes tables to flush pending changes to disk' every flush_time seconds. The means (I think) that some (though I doubt all, given the size of some tables) tables could be completely in memory. Does that imply that if we set this flush_time value to zero (ie no periodic flush to disk), then some of the data will not be committed to disk, and if we had a subsequent power failure, then any data since the last flush would be lost? We have innodb_flush_log_at_trx_commit set to 1. Given the above, is it unwise to drop the periodic flush? The related item: The number of tables in all our databases, including mysql, is 130. What other tables are counted in the opened_tables calculation; does this include tables that may be opened twice under different aliases? Does this include temporary tables (created by MySQL)? Cheers Terry Riley -- MySQL General Mailing List For list archives: http://lists.mysql.com/mysql To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED]