From mboxrd@z Thu Jan 1 00:00:00 1970 From: "NeilBrown" Subject: Re: HugePages for MD's caches -- can this be done in modern kernel? Date: Sat, 7 Feb 2009 07:16:11 +1100 (EST) Message-ID: <39ad4724d7dd3856cec91dc908cd2a7b.squirrel@neil.brown.name> References: <43d009740902061008p968268bj30a03c6f97139bbe@mail.gmail.com> <43d009740902061127p1a8073ax231877c8ed3eeda9@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7BIT Return-path: In-Reply-To: <43d009740902061127p1a8073ax231877c8ed3eeda9@mail.gmail.com> Sender: linux-raid-owner@vger.kernel.org To: for.poige+linux@gmail.com Cc: linux-raid@vger.kernel.org List-Id: linux-raid.ids On Sat, February 7, 2009 6:27 am, Igor Podlesny wrote: > 2009/2/7 NeilBrown : >> On Sat, February 7, 2009 5:08 am, Igor Podlesny wrote: >>> Hi! >>> >>> $(Subj). :-) >>> >>> P. S. Also, I deem there lots of structures in kernel that could >>> benefit >>> from using HugePages. But this probably a q-n for another kernel list. >> >> I have only a vague idea what HugePages are and it is not at all clear >> to >> me why you would want to use them for the MD cache. I assume you are >> > HugePages requires less TBL entries so they better fit for large > buffers, as far as I understand. > > http://unixfoo.blogspot.com/2007/10/hugepages.html > http://linux-mm.org/HugePagesArticles > >> talking about MD/RAID456 as that is the only place we use a cache. >> > Yeah, "stripe_cache_size" setting. >> >> Maybe if you could explain what you are thinking.... >> > Yeah, hope this would suffice. Not really, no. md does not use "large buffers", Just lots of little buffers. NeilBrown