Re: latches or semaphores

Thank you very much.

Yes, I had meant latch posting (_latch_wait_posting).

if _latch_wait_posting=0, waiting process is not posted by holding process even if lath is free. When waiting process wakes up after a timeout, it gets latch. in other words, there is no interprocess communication here.

As you stated, when _LATCH_WAIT_POSTING=1, which is default, waiting process can be posted for library cache and shared pool latches only.

for _LATCH_WAIT_POSTING>1, posting is available for all types of latches.

my question is how interprocess communication (IPC) is implemented in post operations. How does an holding process post another waiting process when it releases latch ? How is a waiting process notified so that it cancels its waiting and gets latch.

Semaphore can not be. Releasing latch by updating shared memory segment is not enough for post operation. Becasue, it's already done for non-posting operations too.

btw,

is Dekker's Algorithm exact algorithm for Oracle before TAS instructions ? or did you mean critical sections were protected like that ?

thanks in advance...

Danisment Gazi Unal
http://www.unal-bilisim.com Received on Thu Nov 15 2001 - 17:57:48 CST