Hello,
I thought Wikimedia uses MySQL replication, not
clustering. MySQL
clustering uses main memory to store tables. Somehow I don't think
the
Wikimedia servers have that much memory.
You're slightly wrong. We have enough memory to store our mysql core
database in memory.
BTW, MySQL 5.1 will finally
support disk-based clustering.
Yes, for non-indexed data.
Even then, there is still the issue of
FULLTEXT indexes in cluster mode. Does the cluster storage engine
support
FULLTEXT indexes?
We don't use FULLTEXT indexes on live site anyway. It's pure InnoDB,
and search is offloaded to Lucene.
Anyway, Cluster is designed to handle zillions of small transactions
per second without data loss, it could work as a solution for session
storage (and it is used that way on several biggish sites).
On the other hand, our access pattern would not like distributed
database that much as there're quite a lot of batch reads.
We already solve the 'clustering' by simply having replicated sets of
database nodes working with different workloads / patterns if needed.
We're quite happy with current setup, I guess ;-)
Domas