From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <446133FD.6090903@yahoo.com.au> Date: Wed, 10 May 2006 10:29:49 +1000 From: Nick Piggin MIME-Version: 1.0 Subject: Re: [RFC][PATCH 1/3] tracking dirty pages in shared mappings -V4 References: <1146861313.3561.13.camel@lappy> <445CA22B.8030807@cyberone.com.au> <1146922446.3561.20.camel@lappy> <445CA907.9060002@cyberone.com.au> <1146929357.3561.28.camel@lappy> <1147116034.16600.2.camel@lappy> <1147207458.27680.19.camel@lappy> <17505.267.931504.918245@wombat.chubb.wattle.id.au> <446101BA.4000208@google.com> In-Reply-To: Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org Return-Path: To: Linus Torvalds Cc: Martin Bligh , Peter Chubb , Peter Zijlstra , Christoph Lameter , Nick Piggin , Andi Kleen , Rohit Seth , Andrew Morton , hugh@veritas.com, riel@redhat.com, andrea@suse.de, arjan@infradead.org, apw@shadowen.org, mel@csn.ul.ie, marcelo@kvack.org, anton@samba.org, paulmck@us.ibm.com, linux-mm List-ID: Linus Torvalds wrote: > On Tue, 9 May 2006, Martin Bligh wrote: > >>Peter Chubb wrote: >> >>>What does this do to performance on TPC workloads? How many extra >>>faults are there likely to be? >> >>They all use large pages anyway ... > > > Yes, but we should really answer that question anyway. > > I don't think anybody has really objected to this patch series, and it > clearly seems to fix something (if nothing else, then Peter's test-program > ;), and for fairly obvious reasons _I_ approve of it. > > But I've been asking people to benchmark it anyway. Becuase it _will_ hurt > users of shared writable mappings, even if it's hopefully less of an issue > today thanks to large pages. > > Now, the fact that it will hurt them is not a total disaster, since we > already know that there are ways to mitigate it. However, it's still true > that we should have the numbers, if only to perhaps be able to say that we > don't care. Those filesystems (shmem, presumably hugetlbfs) that are !mapping_cap_account_dirty should not notice any difference, except perhaps for the several extra branches and bytes of icahce. Or do these databases use regular filesystem backed shared files? -- SUSE Labs, Novell Inc. Send instant messages to your online friends http://au.messenger.yahoo.com -- 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