Very interesting answer, thank you. Partition alignment, and IO performance in general, is something I'm trying to get better grips on. FWIW, we're on AoE 47 (stock Debian kernel), with AoE clients using two AoE devices as a backing device for a RAID1 device. It works well enough, but performance isn't exactly stellar.
I have some questions that would be grateful to get the answer to. * How will using iSCSI change the partition alignment problem ? I'm guessing you have found the difference the only way that works; by trying. It seems unclear why AoE would cause the unaligned writes, so maybe there isn't a good answer to this question. * How do you actually see from the output of iostat that you're writes aren't aligned ? I usually use "iostat -dxm 1" to get an overview, but I don't fully understand all the output. In a non-AoE context, I have a dm RAID6, where I see all the disks and the md device when using iostat. I can there see the md device hitting 100% utilization, while the disk don't go above ~60%. Is that a symptom of unaligned writes ? Thanks for your insightful answers. On Thu, Apr 26, 2012 at 13:02, Tracy Reed <tr...@ultraviolet.org> wrote: > On Wed, Apr 25, 2012 at 01:47:23PM +0200, Lars Täuber spake thusly: >> Tracy Reed <tr...@ultraviolet.org> schrieb: >> > Reasons why I am currently migrating away from AoE to iSCSI (*sigh*): >> > >> > 1. Disk alignment between Xen VMs and the target. >> >> What do you mean by that? > > I mean that when I configure a Xen VM to use an AoE block device I always had > mis-aligned writes. The Xen dom0 has the block device in /dev/etherd and I put > the block device in the Xen VM config file and make it /dev/xvda inside the > VM. > > If I access the device from dom0 everything is fine. Very fast writes, no > misalignment. But accessing the block device from within the VM causes the > problem. This makes no sense to me and I don't see anything that could cause > alignment to change but it clearly did somehow. This got to be very noticeable > performance-wise and when doing a pure-write benchmark while running iostat on > the target I could see lots of reads happening to backfill partial pages due > to > the misaligned write. > > -- > Tracy Reed > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Aoetools-discuss mailing list > Aoetools-discuss@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/aoetools-discuss > -- Vennlig hilsen Torbjørn Thorsen Utvikler / driftstekniker Trollweb Solutions AS - Professional Magento Partner www.trollweb.no Telefon dagtid: +47 51215300 Telefon kveld/helg: For kunder med Serviceavtale Besøksadresse: Luramyrveien 40, 4313 Sandnes Postadresse: Maurholen 57, 4316 Sandnes Husk at alle våre standard-vilkår alltid er gjeldende ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Aoetools-discuss mailing list Aoetools-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/aoetools-discuss
