Re: Large pages

From: Noons <wizofoz2k_at_gmail.com>
Date: Sun, 11 Nov 2012 20:34:06 -0800 (PST)
Message-ID: <10f49835-b336-4d9a-ba93-5e927761a030_at_q5g2000pbk.googlegroups.com>



On Nov 12, 11:24 am, Mladen Gogala <gogala.mla..._at_gmail.com> wrote:

> > In Unix (unless you use intimate shared memory) you get one map per
> > process - so an Oracle system with 1,000 processes would end up using as
> > much memory for maps of the SGA as it would on the SGA itself if it were
> > using standard pages. In Windows Oracle runs as a single process with

Actually, that is not correct. The problem has to do with the memory map for ALL physical memory, the number of virtual memory processes using it is irrelevant.

> Yup, that's precisely why I ask. Windows do not have SYSVR4 IPC
> mechanisms, probably because they do not belong to that family of
> operating systems, and have completely different architecture. I played
> with that once, but I no longer have access to the database I have built,
> so I am asking around.

Should be exactly the same. The problem has to do with mapping virtual to physical memory for very large physical memory sizes. At 4K map per entry, a memory size of say, 50GB will easily force a TLA table that has in excess of 10 million entries, which have to be searched and resolved by the virtual memory firmware as well as the OS, for normal memory management as well as statement execution and branching.
It's got nothing to do with the number of processes. The number of entries in the TLA table maps physical addresses to virtual addresses, how many processes use those virtual addresses is immaterial. The total size of the mapped physical memory does not change because a process is using it or not. Received on Mon Nov 12 2012 - 05:34:06 CET

Original text of this message