4 ms·
Architecture of a site you probably never use: 60 mysql servers (30 master/master pairs) ranging from 2 core machines with 8 GB RAM to 8 core machines with 64 G
by seiji 16y ago
Architecture of a site you probably never use: 60 mysql servers (30 master/master pairs) ranging from 2 core machines with 8 GB RAM to 8 core machines with 64 GB RAM and RamSan PCI SSDs.
Some tables are sharded, others aren't. Some tables are imported into hadoop for hive queries, others aren't. Backups are run using a fragile program called zrm and an army of outsourced DBAs to manage it.
The army of outsourced DBAs also deal with replication failures, figuring out why inexplicable deadlocks happen, tuning every inane innodb configurable, and "rebuilding" servers when mysql decides to slow itself down and yearns for a complete dump/reload of all tables.
Friends don't let friends use mysql.
- rudiger 16y agoInteresting. I've never seen large-scale database deployments use SSDs (neither for the transaction log, nor for the data); I've only seen battery-backup and write-back caches on hard disk drives used.
- zwischenzug 16y agoNor have I. But it may be the future: cf rethinkdb.
- jpitz 16y agoYou will probably start to. The intel 320 series is getting some attention now [1] with a combination of competitive ( for SSD ) price and enough capacitors on board to flush the write cache on a power failure. You'd likely use SSDs first for indexes, then data. Transaction logs are usually a sequential workload and 7200 rpm drives often suffice. [1] http://archives.postgresql.org/pgsql-performance/2011-03/msg00342.php http://archives.postgresql.org/pgsql-performance/2011-03/msg...