From mboxrd@z Thu Jan 1 00:00:00 1970 Subject: Re: hugepage patches References: <20030131151501.7273a9bf.akpm@digeo.com> <20030202025546.2a29db61.akpm@digeo.com> <20030202195908.GD29981@holomorphy.com> <20030202124943.30ea43b7.akpm@digeo.com> <20030203132929.40f0d9c0.akpm@digeo.com> <20030204055012.GD1599@holomorphy.com> <162820000.1044342992@[10.10.2.4]> From: ebiederm@xmission.com (Eric W. Biederman) Date: 04 Feb 2003 05:40:39 -0700 In-Reply-To: <162820000.1044342992@[10.10.2.4]> Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: owner-linux-mm@kvack.org Return-Path: To: "Martin J. Bligh" Cc: William Lee Irwin III , Andrew Morton , davem@redhat.com, rohit.seth@intel.com, davidm@napali.hpl.hp.com, anton@samba.org, linux-mm@kvack.org List-ID: "Martin J. Bligh" writes: > > O.k. Then the code definitely needs to handle shared mappings.. > > Why? we just divided the pagetable size by a factor of 1000, so > the problem is no longer really there ;-) William said one of the cases was to handle massively shared mappings. You cannot create a massively shared mapping except by sharing. Did I misunderstand what was meant by a massively shared mapping? I can't imagine it being useful to guys like oracle without MAP_SHARED support.... > >> Well, in theory there's some kind of TLB benefit, but the only thing > >> ppl really care about is x86 pagetable structure gets rid of L3 space > >> entirely so you don't burn 12+GB of L3 pagetables for appserver loads. > > > > I am with the group that actually cares more about the TLB benefit. > > For HPC loads there is really only one application per machine. And with > > just one page table, the only real advantage is the more efficient use > > of the TLB. > > The reason we don't see it much is that we mostly have P3's which only > have 4 entries for large pages. P4's would be much easier to demonstrate > such things on, and I don't think we've really tried very hard on that with > hugetlbfs (earlier Java work by the research group showed impressive > improvements on an earlier implementation). Cool. I have no doubt the benefit is there. Measuring how large it is will certainly be interesting. Eric -- 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/