Hello, I am sure that this question was answered already but I could not find an answer. Is it possible to force zfs pool to have concatenation not striping or it can''t be specified. Hints would be greatly appreciated. Thanks, Mike.
On Fri, Sep 19, 2008 at 10:31:07AM -0400, Michael Dvinyaninov wrote:> Hello, > > I am sure that this question was answered already but I could not find > an answer. > Is it possible to force zfs pool to have concatenation not striping or > it can''t be specified.No, it can''t. How would having concatenations instead of dynamic striping help you? -- Darren
See http://www.opensolaris.org/jive/thread.jspa?messageID=271983񂙯 The case mentioned there is one where concatenation in zdevs would be useful.
On Mon, Sep 22, 2008 at 01:03:13PM +0200, Nils Goroll wrote:> See > > http://www.opensolaris.org/jive/thread.jspa?messageID=271983񂙯 > > The case mentioned there is one where concatenation in zdevs would beuseful. That case appears to be about trying to get a raidz sized properly against disks of different sizes. I don''t see a similar issue for someone preferring a concat over a stripe. -- Darren
Hi Darren,>> http://www.opensolaris.org/jive/thread.jspa?messageID=271983񂙯 >> >> The case mentioned there is one where concatenation in zdevs would be > useful. > > That case appears to be about trying to get a raidz sized properly > against disks of different sizes. I don''t see a similar issue for > someone preferring a concat over a stripe.I don''t quite understand your comment. The question I was referring to was from someone who wanted a configuration which would optimally use the available physical disk space. The configuration which would yield maximum net capacity was to use concats, so IMHO this is a case where one might want a concat below a vdev. Were you asking for use cases of a concat at the pool layer? I think those exist when using RAID hardware where additional striping can lead to an increase of concurrent I/O on the same disks or I/Os being split up unnecessarily. All of this highly depends upon the configuration. Nils
I actually ran into a situation where I needed to concatenate LUNs last week. In my case, the Sun 2540 storage arrays don''t yet have the ability to create LUNs over 2TB, so to use all the storage within the array on one host efficiently, I created two LUNs per RAID group, for a total of 4 LUNs. Then we create two stripes (LUNs 0 and 2, 1 and 3) and concatenate them. This way the data is laid out contigously on the RAID groups. -Aaron On Mon, Sep 22, 2008 at 11:56 PM, Nils Goroll <slink at schokola.de> wrote:> Hi Darren, > > >> http://www.opensolaris.org/jive/thread.jspa?messageID=271983񂙯 > >> > >> The case mentioned there is one where concatenation in zdevs would be > > useful. > > > > That case appears to be about trying to get a raidz sized properly > > against disks of different sizes. I don''t see a similar issue for > > someone preferring a concat over a stripe. > > I don''t quite understand your comment. > > The question I was referring to was from someone who wanted a configuration > which would optimally use the available physical disk space. The > configuration > which would yield maximum net capacity was to use concats, so IMHO this is > a > case where one might want a concat below a vdev. > > Were you asking for use cases of a concat at the pool layer? > > I think those exist when using RAID hardware where additional striping can > lead > to an increase of concurrent I/O on the same disks or I/Os being split up > unnecessarily. All of this highly depends upon the configuration. > > Nils > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20080923/451c2ff6/attachment.html>
Hello Aaron, Tuesday, September 23, 2008, 8:24:36 AM, you wrote: > I actually ran into a situation where I needed to concatenate LUNs last week. In my case, the Sun 2540 storage arrays don''t yet have the ability to create LUNs over 2TB, so to use all the storage within the array on one host efficiently, I created two LUNs per RAID group, for a total of 4 LUNs. Then we create two stripes (LUNs 0 and 2, 1 and 3) and concatenate them. This way the data is laid out contigously on the RAID groups. Depending on your pattern use you could get sub-optimal performance. If you create smaller raid grups and match one lun to a given raid group and then use a zfs striping across luns you probably will get better performance. -- Best regards, Robert mailto:milek@task.gda.pl http://milek.blogspot.com _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
On Tue, Sep 23, 2008 at 08:56:39AM +0200, Nils Goroll wrote:>> That case appears to be about trying to get a raidz sized properly >> against disks of different sizes. I don''t see a similar issue for >> someone preferring a concat over a stripe. > > I don''t quite understand your comment. > > The question I was referring to was from someone who wanted a > configuration which would optimally use the available physical disk > space. The configuration which would yield maximum net capacity was to > use concats, so IMHO this is a case where one might want a concat below a > vdev.I assumed that was only necessary because they were trying to use a raidz. I don''t see why a plain stripe would not yield the maximum capacity.> Were you asking for use cases of a concat at the pool layer? > > I think those exist when using RAID hardware where additional striping > can lead to an increase of concurrent I/O on the same disks or I/Os being > split up unnecessarily. All of this highly depends upon the > configuration.No, not necessarily. I can think of a couple of things where it might be useful. I was asking for the OP''s particular need. -- Darren