SUMMARY: multi-terabyte disk support

Executive Summary:  Solaris 9 version 8/03 has this functionality:  It can
handle "disks" up to 16TB in size, and filesystems of the same size.  This
functionality requires running the 64-bit kernel.

Thanks to several people for responding:
Justin Stringfellow
Casper Dik
Jay Lessert
Hichael Morton
Grzegorz Bakalarski
Sandeep Ghodke
(I hope procmail/bogofilter didn't eat anyone else's response)

Thanks also to lots of people for letting me know that they are out of the
office.  I (well, OK, procmail) actually got the first "Out of Office" before
the original message came back from the list.

I delayed sending this summary because I wanted to actually test the patches
in question.  I currently have a 3TB filesystem attached to our backup 
server, and our databases are being copied quickly from disk to disk.

The original question: (more info below)

On Thu, Aug 07, 2003 at 02:44:31PM -0400, Joe Moore wrote:
> I have a large ATA RAID array (Nexsan ATAboy2f) attached to my backup
> server (SunFire V880 running Solaris 8).  For various reasons, I'd like
> to be able to present the entire array (about 3TB) as a single filesystem.
> 
> According to Sun, all versions of Solaris are limited to 1TB in a single
> filesystem.  There is a patch for Solaris 9, however (114129-01 and others)
> which seem to bypass this limitation:
> 
> (From the Free Patch Description of 114129-01)
> NOTE:  To get the complete Multi-terabyte feature, please install the following
>         patches:
>  
>         113277-06 (or newer) sd & ssd patch
>         114127-01 (or newer) abi_libefi.so.1 patch
>         114128-01 (or newer) sd_lun patch
>         114131-01 (or newer) libadm.so.1 patch
>         113981-02 (or newer) devfsadm patch
>         114132-01 (or newer) fmthard patch
>         113072-03 (or newer) format patch
>         114369-01 (or newer) prtvtoc patch
>         113049-01 (or newer) luxadm & liba5k.so.2 Patch
> 
> Does anyone have experience with this patch set, or with very large 
> filesystems in general?
> 
> I've looked at vxfs, but I must explore lower-cost options before a purchase
> will be approved.

These patches, which can be applied to Solaris 9, are only half of the picture.
These allow the system to access a 16TB "disk".  There are also patches
required for the mkfs, fsck, etc. commands.  I was not able to find the patches
for these commands, only references to the new -T option for mkfs.
I ended up doing an upgrade install to the Solaris 9 08/03 environment.

UFS under Solaris is a mature, stable filesystem which has been optimized
over the years to give very good performance.  For large UFS filesystems, 
you need to use logging to prevent multi-day fsck operations at boot time.
Fortunately, when created with the -T option to mkfs/newfs, logging is the
default.

I had to create new filesystems on the disk array.  In retrospect, this makes
sense.  Apparently, the multi-terabyte feature is implemented by changing the
addressing model of disk blocks in the filesystem.  With older UFS filesystems,
(superblock magic FS_MAGIC) offsets were stored in terms of sectors (512 
bytes), in a 32-bit field, which limits the largest offset to 1099511627776
bytes from the start of the disk.  With the multi-terabyte UFS filesystem, 
(superblock magic MTB_UFS_MAGIC), the offset is stored in terms of the block 
size (8192 bytes, a 16x larger unit).  ufs(7FS) documents a lot of this,
and $INCLUDE/sys/fs/ufs_{fs,inode}.h are also useful.

Thanks again for all the help,
--Joe
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers