Adfs Filesystem?? What's That [Re: DoveCot vs Cyrus-Imapd Performance]

Aleksandar Milivojevic amilivojevic at pbl.ca
Mon Jan 17 17:27:35 UTC 2005


Chris Adams wrote:
> UFS is only supported in read-only mode in a TruCluster setup, while
> AdvFS is fully cluster-aware and supports full POSIX semantics in a
> cluster environment (well, except for the times is doesn't :-) ).  It
> doesn't include full LVM functionality (like it can't do RAID except for
> striping, which is done at the file level instead of the block level),
> but it does integrate many of the useful parts.  You can do snapshots at
> the filesystem level instead of the block level (which means you don't
> have to leave disk space unallocated to do snapshots).  AdvFS can also
> have multiple filesets in a single file domain, all sharing space
> (optionally with quotas on each fileset).

Yup, RAID needed to be done on separate level, but I never missed that 
one.  If you had TruCluster, you'd probably have some external hardware 
RAID to take care of that.  And I really liked the integration of LVM 
features with file system that advfs had.  It simplified things.  You 
want to move bunch of file systems to larger disk.  Add it to advfs 
domain, delete old disk from domain, and voila you are done.  No manual 
file system resizing and stuff.

And yeah, the ability for logical volumes (file sets in advfs speak) to 
share all the space in volume group (file domain in advfs speak) instead 
of allocating fixed amount to each volume was really great.  We are not 
going to see this in Linux unless ext3 (or some other fs type) becomes 
LVM aware.

BTW, Sun DiskSuite (which is basically software RAID implementation, not 
LVM) can also do snapshots without reserving disk space for it.  With 
DiskSuite, submirror can be either "offlined" or "detached".  If you 
detach/attach submirror, it will be resynced from scratch.  However, if 
you offline submirror, when you online it again, system will sync only 
changed blocks.  So I guess that feautre could be more or less easilly 
implemented in Linux.  Not sure where it keeps what blocks need updating 
(256 kB of data per 1 GB of disk space).  Kernel memory?  Swappable 
memory?  RAID metadata stored on disk (such as the space at the end of 
partition that md uses on Linux)?

Heh, this thread really got wild.  From Dovecot vs Cyrus, we got to the 
kernel wishlist ;-)

> AdvFS really rocks; I will miss it (and TruCluster) when the Alpha
> disappears.  Thanks HPAQ! :-(

Yup.  Mee too :-(

-- 
Aleksandar Milivojevic <amilivojevic at pbl.ca>    Pollard Banknote Limited
Systems Administrator                           1499 Buffalo Place
Tel: (204) 474-2323 ext 276                     Winnipeg, MB  R3T 1L7




More information about the users mailing list