From: aoki@postgres.Berkeley.EDU (Paul M. Aoki) To: kskim@hyowon.pusan.ac.kr (kim kyongsok) Subject: Re: some questions. In-reply-to: Your message of Wed, 15 Sep 1993 12:14:26 +0900 <9309150314.AA21661@hyowon.pusan.ac.kr> Date: Tue, 14 Sep 93 21:06:50 -0700 Sender: aoki X-Mts: smtp kskim@hyowon.pusan.ac.kr (kim kyongsok) writes: > 1. postgres uses shared memory. does shared memory reside on real > memory or on virtual memory (and thus paged). the system v shared memory interface works with virtual addresses. you attach a segment to your virtual address space with shmat(3). shared memory segments can be paged (SHM_UNLOCK) or unpaged (SHM_LOCK). this is supposed to be controlled by shmctl(3). postgres never does SHM_LOCK. i think many versions of unix default to SHM_LOCK behavior -- sunos3, for example. but sunos does not document SHM_LOCK. on ultrix, you *can* call shmctl(SHM_LOCK) explicitly but you have to be uid 0.. hpux has similar restrictions. so, since no version of unix really lets you set page/no-page as a regular user, postgres has to settle for whatever the OS provides by default. (or run as root. probably a bad idea :-) you can look at the man pages, and the kernel source if you're really interested. if you want to publish this somewhere don't take my word for it -- check the kernel source. > 2. is postgres's buffer for database pages in shared memory? yes. > 2. a related question. does postgres's buffer for database pages > reside on real memory or on virtual memory? "related"? this is the same as question 1 + question 2.. -- Paul M. Aoki | CS Div., Dept. of EECS, UCB | aoki@postgres.Berkeley.EDU | Berkeley, CA 94720 | ...!uunet!ucbvax!aoki