Hi list, I''ve just implemented a lustre system which has 4 OSS and 32 SSD hard disk. But, I got a big problem: lustre doesn''t support new kernel. That means, no SSD TRIM command is supported. So, when write files, the system is too slow, and sometime crash Any idea? Thanks. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.lustre.org/pipermail/lustre-discuss/attachments/20101123/5684ee25/attachment.html
On Tue, Nov 23, 2010 at 03:12:41PM +0700, sd a wrote:> I''ve just implemented a lustre system which has 4 OSS and 32 SSD hard disk. > But, I got a big problem: lustre doesn''t support new kernel. That means, no > SSD TRIM command is supported.1.8.5 (should be released soon) will support the SLES11 SP1 kernel (based on 2.6.32) which supports the discard/TRIM commands thanks to the "discard" mount option. However, we have not tested it, so i cannot really tell whether or not it works well with ldiskfs/lustre. Johann
On 2010-11-23, at 01:29, Johann Lombardi wrote:> On Tue, Nov 23, 2010 at 03:12:41PM +0700, sd a wrote: >> I''ve just implemented a lustre system which has 4 OSS and 32 SSD hard disk. But, I got a big problem: lustre doesn''t support new kernel. That means, no SSD TRIM command is supported. > > 1.8.5 (should be released soon) will support the SLES11 SP1 kernel (based on 2.6.32) which supports the discard/TRIM commands thanks to the "discard" mount option. However, we have not tested it, so i cannot really tell whether or not it works well with ldiskfs/lustre.Note that testing of the "discard" mount option (by other ext4 filesystem developers, not related to Lustre) has shown that it can significantly hurt the io performance due to the slowness of the TRIM commands, and the kernel implements TRIM like a barrier so it slows down all of the IO. Cheers, Andreas -- Andreas Dilger Lustre Technical Lead Oracle Corporation Canada Inc.
as far as I know, since 2.6.33, linux supports ''full'' TRIM command, the IO problem is solved On Tue, Nov 23, 2010 at 4:07 PM, Andreas Dilger <andreas.dilger at oracle.com>wrote:> Note that testing of the "discard" mount option (by other ext4 filesystem > developers, not related to Lustre) has shown that it can significantly hurt > the io performance due to the slowness of the TRIM commands, and the kernel > implements TRIM like a barrier so it slows down all of the IO. > > Cheers, Andreas > -- > Andreas Dilger > Lustre Technical Lead > Oracle Corporation Canada Inc. > >-------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.lustre.org/pipermail/lustre-discuss/attachments/20101123/5fdc4b48/attachment.html
On Tue, Nov 23, 2010 at 05:09:22PM +0700, sd a wrote:> as far as I know, since 2.6.33, linux supports ''full'' TRIM command, the IO > problem is solvedIndeed, i *think* this patch is the one: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=18f0f97850059303ed73b1f02084f55ca330a80c However, it is surprising that neither RedHat nor SuSE have backported this patch to their 2.6.32-based kernel. Johann