From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Wed, 31 Aug 2005 08:39:13 -0700 From: "Martin J. Bligh" Reply-To: "Martin J. Bligh" Subject: Re: [PATCH 1/1] Implement shared page tables Message-ID: <19490000.1125502753@[10.10.2.4]> In-Reply-To: References: <7C49DFF721CB4E671DB260F9@[10.1.1.4]><1125489077.3213.12.camel@laptopd505.fenrus.org><16640000.1125498711@[10.10.2.4]> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline Sender: owner-linux-mm@kvack.org Return-Path: To: Hugh Dickins Cc: Arjan van de Ven , Dave McCracken , Andrew Morton , Linux Kernel , Linux Memory Management List-ID: >> They're incompatible, but you could be left to choose one or the other >> via config option. > > Wouldn't need config option: there's /proc/sys/kernel/randomize_va_space > for the whole running system, compatibility check on the ELFs run, and > the infinite stack rlimit: enough ways to suppress randomization if it > doesn't suit you. Even better - much easier to deal with distro stuff if we can do it at runtime. >> 3% on "a certain industry-standard database benchmark" (cough) is huge, >> and we expect the benefit for PPC64 will be larger as we can share the >> underlying hardware PTEs without TLB flushing as well. > > Okay - and you're implying that 3% comes from _using_ the shared page > tables, rather than from avoiding the fork/exit overhead of setting > them up and tearing them down. And it can't use huge TLB pages > because... fragmentation? Yes - as I understand it, that was a straight measurement with/without the patch, and the shmem segment was already using hugetlb (in both cases). Yes, I find that a bit odd as to why as well - they are still trying to get some detailed profiling to explain. M. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org