You are mostly right, hence the addendum. As you can see there is some effect in the first benchmark. As far as I know now, query_cache is beneficial for InnoDB, but only in a situation with a low number of threads and the innodb buffer pools is a much better use of available memory than the query cache.
]]>key_buffer and query_cache should not have any effect, other than controlling how much memory is unnecessarily wasted. These settings are for MyISAM.
]]>Can you post your 50 thread script/command please?
Thanks
]]>I can’t speak for Mo, but for Observu we’ve selected DIY MySQL in combination with Galera for replication. The main reason to let RDS pass is because there is no easy way to get your data out of there without downtime. (mysqldump isn’t that much fun for big databases) Another one is that failover in the Multi-AZ setup is reported to take a few minutes (due to DNS caching, etc) Using Galera, we can make multiple servers known to our app and we can do failover in the application layer if needed.
]]>Which one did you end up going with? How did it go?
]]>