>>>>> "Mauricio" == Mauricio Breternitz <mbjsql@stripped> writes:
Mauricio> As Sasha said, intra-query parallelization is a VERY HARD TASK.
Mauricio> However, even an SMP could take a throughput approach and run
Mauricio> several copies of the MySQL server (would get some parallelism
Mauricio> in the query optimization, etc, stages) before hitting the file system.
Mauricio> Each processor would execute one query completely. Inter-processor
Mauricio> locks are needed to protect the next query from trying to hit
Mauricio> the table before the previous query is finished in another processor
Mauricio> I'm not on an SMP, however the processors share access to the disks.
Mauricio> The throughput approach is still feasible.
Mauricio> Any suggestions at which files should I look into as reasonable place to put
Mauricio> inter-processor locks ?
Mauricio> (I'm currently delving in the sql/*.cc files and looking at things
Mauricio> like open_and_lock_tables ),
If you are just searching after locks, then the place to do this is
When a thread asks for access to a table, it always calls
thr_lock() to get the lock and will execute thr_unlock() when it's ready.
This is even true for BDB tables, but in this case thr_lock() allows
multiple read and write locks at the same time.