How is synchronisation performed between classic architecture server
processes?

I assume that there are short locks on pages in the database file. Ie:
acquire write lock, read/modify page, write page, release lock, or acquire
read lock, read page, release lock. This would avoid the problems of
concurrent updates to the same page and reads of partially written pages on
operating systems without atomic writes (eg. some System V.2 systems).

If this locking exists, how is it implemented? If not, how is page level
synchronisation performed? Does the implementation vary significantly by
platform?

I assume higher level isolation is totally taken care of by the protocols
around the on disk structure with transaction numbers, and so needs no
separate IPC channel.

Jan Mikkelsen
janm@...