> :Thanks Dillon for your detailed informative reply :-)
> :
> :I am actually trying to upgrade my backup server which currently runs
> :OpenBSD and backuppc from 2 120GB SATA drives on RAID1 usinf Raidframe
> :to 2 500 GB SATA drives running on RAID1.
> :
> :Do you think I can safely use HammerFS and dragonfly for this?
> :Most of my search in the net gave me results that HammerFS is not yet
> :ready for production.
> :My concern is because this is a backup server.
> 
>    I'm using a 750G HAMMER filesystem on the DragonFly backup box.  HAMMER
>    is very well suited towards this sort of operation.  The only downside
>    is directory scan performance isn't as high as it could be.  It gets
>    better as one gets deeper in the directory structure, though.
> 
>    In anycase, using HAMMER for the backup drive is a near perfect match
>    because you can let HAMMER do the snapshotting for you.  Basically
>    Our backup box NFS-ro-mounts the other 6 machines and uses cpdup to
>    do a nightly backup.  Everything is self contained in the backup box
>    itself so there is no impact on the production system if the backup box
>    goes down.  I don't have to use hardlink tricks or renames or anything
>    else, I just cpdup nightly and then use HAMMER to create a snapshot.
> 
>    The offsite runs weekly and uses the latest snapshot directory on the
>    LAN backup box so it has a stable copy as the offsite source.
> 
>    I can hold about 120 days worth of daily snapshots on the backup box
>    using this method, and over a year's worth of weekly backups on the
>    offsite box.
> 
>    Setting up the snapshotting is fairly straight-forward.  Basically
>    the daily cron runs 'hammer cleanup' which sets up a <fs>/snapshots/config
>    file automatically, you just have to be careful that it doesn't interfere
>    with whatever custom snapshotting scheme you have.  If you put the
>    snapshots in the <fs>/snapshots directory but you want to generate your
>    own snapshot softlinks instead of having hammer cleanup do it (so they
>    are synchronized with the backup), then edit the <fs>/snapshots/config
>    file as follows:
> 
>        snapshots 0d 120d
>        prune     1d 60m
>        reblock   1d 60m
>        recopy    30d 60m
> 
>        (snapshots 0d Xd -- means the cron hammer cleanup doesn't create
>        any snapshots, since you are creating custom ones.  The rest of
>        the parameters are just beefed up defaults since the backup drive
>        is very large).
> 
>    Basically longer pruning and reblocking times then the default, turn off
>    hammer cleanup's own snapshotting (0d on the snapshots config file),
>    but do tell it to delete snapshots over 120 days old (adjust as required
>    once the disk starts to fill up, so it doesn't become too full).
> 
> :Also the system is amd64. Can I safely install the i386 port of
> :dragonfly on it and expect it to work properly? ( dmesg of the machine
> :at the end. )
> 
>    Not sure what you mean there.  Yes, you want to install the main
>    32 bit distribution.  Our 64 bit distribution is still under development.
>    32 bit boot works fine on amd64.
> 
> :And is the best thing to do now for raid dragonfly+HammerFS+Vinnum ?
> 
>    Not for a backup box.  For a backup box I'd just use a single large
>    hard drive, preferably one in a hot-swap enclosure that I can just
>    pull out and take with me in an emergency.  I then also have an off-site
>    backup box.  As long as there are two backup machines (one on-site, one
>    off-site), there is no real need to set up RAID or same-box redundancy.
> 
>    The storage requirements are too high... it's the production machines
>    that might need the RAID, since they are production machines.  The
>    backup box is not a time-critical component, meaning that you want it
>    to be around but it doesn't create a disaster if it goes down for a
>    day or two.
> 
>    Theoretically I guess you could use vinum and set up a RAID-1 with
>    two drives, but I don't consider that any more reliable then one drive.
>    A vinum'd setup can be used for the root mount with appropriate kenv
>    variables set in /boot/loader.conf.
> 
>    What I do do is put the /backup partition on its own private drive and
>    leave the rest of the backup machine's system on a smaller drive.
>    That way I can umount the backup drive, pull it out, throw a new one,
>    and put the old one on a shelf somewhere as an archive.
> 
> :Is HammerFS available as choice during the installation in the current rele=
> :ase?
> :
> :And Finally should I go for the release or the current?
> :
> :Thanks
> :
> :Siju
> 
>    You want to go with current.
> 
>    HAMMER is an installation choice but the setup we really recommend is
>    not yet available on the installer.  The best setup to have (in my
>    view) is a BOOT+HAMMER setup.  A small UFS BOOT partition 'a', say
>    256M, swap on 'b', and the rest of the disk would be a HAMMER FS on
>    'd' ('d' becomes the root fs).  You mount the boot filesystem on 'a'
>    under the root fs as /boot.  Setting it up is fairly straight-forward,
>    the only gotcha is how to tell the kernel where the root filesystem is
>    since it isn't on 'a'.  You do that by adding a line to /boot/loader.conf
>    that looks something like:
> 
>        vfs.root.mountfrom="hammer:da0s1d"
> 
>    For a multi-disk setup making /boot redundant is a bit more difficult.
>    What I recommend is simply creating a non-RAID /boot on every drive so
>    the BIOS can boot from any one of them.  If it is really important
>    then put the base system on a SSD and boot from there.  SSD's have the
>    same MTBF as the motherboard so there isn't any need, or point, in
>    RAIDing them.  SSDs are actually quite reasonably priced for the low
>    capacities needed for a base system partition.
> 
>    I don't like the fake-raid BIOS solutions.  I really don't like them.
>    They are not necessarily compatible between motherboards so just being
>    able to pull the drives doesn't mean you can easily recover the data
>    on another machine.  When it comes down to it, a solid state drive (SSD)
>    is the best way to boot a machine that absolutely needs to boot reliably.
> 
>    Once you get past the basic system boot you have a lot more options, and
>    intelligence, with regards to mounting the other drives.
> 
>                                                -Matt

Great to see Dillon replying with such detailed information. HammerFS is also 
being 
ported to Linux via FUSE.

 
_______________________________________________
bsd-india mailing list
[email protected]
http://www.bsd-india.org/mailman/listinfo/bsd-india

Reply via email to