@ypercube For the indexes, it would only be stores of
PRIMARY
s; at 1 million
BIGINT
s, it'd be only ~16mb. For rows, infrequent or unused ones would be removed from memory based upon open user sockets and time limits, with time limits on inactive connections. Only text will be buffered, definitely not images which require an extra user step to extract from the database. All site data will be stored in the database, will be written before updating c++ memory then users, and will be heavily normalized to allow for wiki-type editing, but the
FOREIGN
s will never be
UPDATE
d. —
Gracchus 7 mins ago