Hi Ho, Two questions for the zfs developers: 1. Are there any plans for a Shared zfs akin to Shared QFS? The specific function I''m thinking is OracleRAC on SunCluster .. Not necessarily the performance of QFS (with split metadata) but rather the ability to import a zfs pool onto more than one system at the same time, a la CVM. In that way I can use filesystems for RAC, and not bother with a zillion volumes .. 2. zfs can do block volumes, what about raw? Are there any plans to introduce that functionality? [ Just in case option 1. isn''t on the cards ;-) ] Given zfs''s self-healing ablities and RAID-Z; rather than using a traditional LVM to provide raw volumes for a database or application, I''d love to be able to use zfs to provide the LVM layer .. and get all my snapshots, clones, self-healing etc. for free. Just curious .. Thanks, gunther :) Sydney, AU
> 2. zfs can do block volumes, what about raw? > Are there any plans to introduce that functionality?No needs for plan. Each zvol has both block and char device zulu:/home/imp# file /dev/zvol/*/*/* /dev/zvol/dsk/test/vol0: block special (182/1) /dev/zvol/rdsk/test/vol0: character special (182/1) Just like any other disk/volume Cyril> [ Just in case option 1. isn''t on the cards ;-) ] > > Given zfs''s self-healing ablities and RAID-Z; rather than using a > traditional LVM to provide raw volumes for a database or application, > I''d love to be able to use zfs to provide the LVM layer .. and get all > my snapshots, clones, self-healing etc. for free. > > Just curious .. > > Thanks, > > gunther :) > Sydney, AU > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss >
On Thu, Dec 01, 2005 at 01:11:52PM +1100, Gunther Feuereisen wrote:> > 2. zfs can do block volumes, what about raw? > Are there any plans to introduce that functionality? > [ Just in case option 1. isn''t on the cards ;-) ] > > Given zfs''s self-healing ablities and RAID-Z; rather than using a > traditional LVM to provide raw volumes for a database or application, > I''d love to be able to use zfs to provide the LVM layer .. and get all > my snapshots, clones, self-healing etc. for free.I''m not sure quite what you''re asking; zvols do provide rdsk devices as well: # zfs create -V 100M pool/volume # zfs list pool/volume NAME USED AVAIL REFER MOUNTPOINT pool/volume 97.5K 7.66G 97.5K - # ls -lL /dev/zvol/*/pool/volume brw------- 1 root sys 286, 1 Nov 30 18:03 /dev/zvol/dsk/pool/volume crw------- 1 root sys 286, 1 Nov 30 17:55 /dev/zvol/rdsk/pool/volume # cat /etc/name_to_major > /dev/zvol/rdsk/pool/volume # Does that answer your question? Cheers, - jonathan -- Jonathan Adams, Solaris Kernel Development
> 1. Are there any plans for a Shared zfs akin to Shared QFS? > > The specific function I''m thinking is OracleRAC on SunCluster .. > > Not necessarily the performance of QFS (with split metadata) but > rather the ability to import a zfs pool onto more than one system > at the same time, a la CVM.SunCluster support for HA-ZFS should be finished sometime this month. We''re also looking at closer integration with third-party clustering products; more on that as it unfolds.> Given zfs''s self-healing ablities and RAID-Z; rather than using a > traditional LVM to provide raw volumes for a database or application, > I''d love to be able to use zfs to provide the LVM layer .. and get all > my snapshots, clones, self-healing etc. for free.Yep -- or you can use a file. Either way, the one non-automatic thing you''ll want to do is specify the volume''s block size, or the filesystem''s record size, to match the database record size. Otherwise performance will pretty much reek. Jeff
> -----Original Message----- > From: Jonathan Adams [mailto:jonathan.adams at sun.com] > Sent: Thursday, 1 December 2005 18:45 > To: Gunther Feuereisen > Cc: zfs-discuss at opensolaris.org > Subject: Re: [zfs-discuss] Shared zfs + Raw zfs volumes > > > On Thu, Dec 01, 2005 at 01:11:52PM +1100, Gunther Feuereisen wrote: > > > > 2. zfs can do block volumes, what about raw? > > Are there any plans to introduce that functionality? > > [ Just in case option 1. isn''t on the cards ;-) ] > > > > Given zfs''s self-healing ablities and RAID-Z; rather than using a > > traditional LVM to provide raw volumes for a database or > application, > > I''d love to be able to use zfs to provide the LVM layer > .. and get all > > my snapshots, clones, self-healing etc. for free. > > I''m not sure quite what you''re asking; zvols do provide rdsk > devices as well: > > # zfs create -V 100M pool/volume > # zfs list pool/volume > NAME USED AVAIL REFER MOUNTPOINT > pool/volume 97.5K 7.66G 97.5K - > # ls -lL /dev/zvol/*/pool/volume > brw------- 1 root sys 286, 1 Nov 30 18:03 > /dev/zvol/dsk/pool/volume > crw------- 1 root sys 286, 1 Nov 30 17:55 > /dev/zvol/rdsk/pool/volume > # cat /etc/name_to_major > /dev/zvol/rdsk/pool/volume > #Thanks Jonathan (and Cyril Plisko [cyril.plisko at gmail.com] in another reply) for pointing out that the raw devices exist. The question came out after reading the current zfs man page, which states: "The zfs command configures ZFS datasets within a ZFS storage pool, as described in zpool(1M). A dataset can be one of the following: filesystem A standard POSIX file system. ZFS file systems can be mounted within the standard file system namespace and behave like any other file system. volume A logical volume exported as a block device. This type of dataset should only be used under special circumstances; file systems are typically used. Volumes cannot be used in a local zone. snapshot A read-only version of a file system or volume at a given point in time. It is specified as filesystem at name or volume at name." As volumes as block devices were the only ones explicitly stated as being a possible dataset, I thought there might be a reason that I could not use a raw volume. Could anyone clarify? Thanks, gunther :)
Hi Jeff,> -----Original Message----- > From: Jeff Bonwick [mailto:bonwick at zion.eng.sun.com] > Sent: Thursday, 1 December 2005 20:40 > To: zfs-discuss at opensolaris.org; gunther at gfh.com.au > Subject: Re: [zfs-discuss] Shared zfs + Raw zfs volumes > > > > 1. Are there any plans for a Shared zfs akin to Shared QFS?[ Snip .. ]> SunCluster support for HA-ZFS should be finished sometime this month.Cool :)> We''re also looking at closer integration with third-party > clustering products; more on that as it unfolds.:) :) :)> > Given zfs''s self-healing ablities and RAID-Z; rather than using a > > traditional LVM to provide raw volumes for a database or > application, > > I''d love to be able to use zfs to provide the LVM layer > .. and get all > > my snapshots, clones, self-healing etc. for free. > > Yep -- or you can use a file.Yes; just some DBAs resist using files .. because they''ve always used raw devices and are reluctant to change .. it''d be nice to know if at the storage pool layer, I can use zfs magic and then provide them raw volumes or a nice zfs dataset, as needed :)> Either way, the one non-automatic thing you''ll want to do is > specify the volume''s block size, or the filesystem''s record > size, to match the database record size. Otherwise > performance will pretty much reek.Yup .. that''s what I was thinking too .. I saw the other message about setting the blocksize (yesterday''s posts) .. Thanks for your reply! gunther :)
Gunther Feuereisen wrote:> Two questions for the zfs developers:how about an answer from the peanut gallery? ;-)> 1. Are there any plans for a Shared zfs akin to Shared QFS? > > The specific function I''m thinking is OracleRAC on SunCluster .. > > Not necessarily the performance of QFS (with split metadata) but > rather the ability to import a zfs pool onto more than one system > at the same time, a la CVM.Nit: CVM is not a file system. [drifting off-topic] Our "Red October" announcement yesterday might have an interesting affect on the CVM business. SVM is open source and does support RAC already, but Sun has charged a fee for the RAC feature (don''t ask me). Meanwhile, for those DBAs who insist on a file system for RAC, QFS has proven to be a very reliable solution. It will be interesting to see what happens as Red October rolls out.> In that way I can use filesystems for RAC, and not bother with a > zillion volumes ..<wet-blanket> You are assuming that Oracle will support this. Historically, there have been issues with what Oracle will support from the platform vendors. Do not assume that even if it works, Oracle will support it. </wet-blanket> Also, consider that Oracle is pushing ASM rather strongly and we should expect that to continue into the future. If you consider ZFS as a method of managing data with better context, then ASM is Oracle''s equivalent. Over time, it will be increasingly difficult for ZFS to be superior to ASM for running the special-purpose application known as the Oracle database. I don''t think this is a bad thing, it is a natural evolution. Of note, our recent, record-breaking TPC-H benchmark uses ASM. http://www.tpc.org/tpch/results/tpch_result_detail.asp?id=105113001 http://biz.yahoo.com/prnews/051130/sfw034.html?.v=37 -- richard
On Nov 30, 2005, at 18:11, Gunther Feuereisen wrote:> 1. Are there any plans for a Shared zfs akin to Shared QFS?Yes, there are, but we have not yet done any real work in that area. I''m glad to see that there is interest in us doing this. --Ed -- Ed Gould Sun Microsystems File System Architect Sun Cluster ed.gould at sun.com 17 Network Circle +1.650.786.4937 MS UMPK17-201 x84937 Menlo Park, CA 94025