Actually, the db we use had similar problems. Schema changes require at least _some_ downtime, if only to lock the tables. The trouble is less the number of tra

Actually, the db we use had similar problems. Schema changes require at least _some_ downtime, if only to lock the tables. The trouble is less the number of transactions than the linkage between each transaction and other transactions' liabilities in the system. In effect, serious DB's are in effect in-memory updates (some of our bigger customers run with no logical log write to disk on commit (really!)) running on 128-core Sun boxes (principal and failover). You can get through a lot of tps and the saving on dev cost and maintenance through this monolithic simplicity is great.