From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <000b01c36ca9$7bc29b50$c5000a0a@TTHENRIC2> From: "Henric Andersson" References: <000e01c36bf9$e5f82720$6402000a@gasaraki> <20030827115857.GE23955@www.13thfloor.at> <001001c36c93$9336a490$c5000a0a@TTHENRIC2> <20030827121655.GB26817@www.13thfloor.at> Subject: Re: [linux-lvm] Poor performance with LVM 1.0.7 MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Sender: linux-lvm-admin@sistina.com Errors-To: linux-lvm-admin@sistina.com Reply-To: linux-lvm@sistina.com List-Help: List-Post: List-Subscribe: , List-Unsubscribe: , List-Archive: Date: Wed Aug 27 10:58:06 2003 List-Id: Content-Type: text/plain; charset="us-ascii" To: linux-lvm@sistina.com > I asked this, because I reported several times that > lvm does have some I/O bottleneck/issue ... > > I had a similar problem with RAID5 where I could > get 16Mb/s from each disk and about 30Mb/s from > the RAID5 (4 disks) but only 18Mb/s from the lvm > ontop of that raid ... I've seen some evidence of that too on the net, but noone seems to have any answer. Did a test now with bonnie on the LV setup with a 512MB file, and this is the result: -------Sequential Output-------- Char: 17159K/s (97% CPU) Block: 21966K/s (7% CPU) Rewrite: 21052K/s (8% CPU) ---Sequential Input-- Char: 10879K/s (54% CPU) Block: 18166K/s (2% CPU) So, clearly, this should be enough to feed samba the 9.5MB/s that all the other non-lvm disks can do. But apparently not. Is there perhaps some strange configuration of samba with LVM that actually makes it slower? (Due to buffers, and read sizes?) > > Even tried doing: > > > > time cat "somefile on the LV set" > /dev/null > > > > Which performed equally bad (ie, about 4-5MB/s). > > Doing the same on the /dev/hda1 drive gives ALOT > > better performance (more in the line of what hdparm -Tt reports) > > this was, on one of my systems, the reason why > I removed the lvm and used the plain disk ... No good in my case, I need to be able to extend the filesystem and thus, I need LVM. > Heinz suggested to use the new LVM2 anyway, and maybe > this is an option for you, so you could try this too > (if you need patches for LVM2, hopefully they will be > available for 2.4.22 this evening ...) Isn't LVM2 just a new bunch of userland tools (+ a devicemapper) ? Or did I miss something? /Henric