hardware recommendations for MURDER?

[Dale Ghent]

On Jul 20, 2006, at 6:11 PM, Vincent Fox wrote:

> I have to read up on ZFS though, haven't really tinkered with it.  
> If I am
> just running a conventional bunch of backends with local-attached  
> disks of
>  normal sizes, is  ZFS still of benefit to me?

ZFS is certainly a benefit even if you're using it to mirror two 50GB  
slices of 200GB drives.

1) Generally speaking, ZFS combines what you would do with a  
traditional LVM and the file system itself. It embodies the "pools"  
concept where, for example, you take a bunch of disks and either  
mirror (RAID-10/1) them or use RAID-Z[1] to create a member of that  
pool. This pool and its member(s) represent one full ZFS filesystem  
that you can then carve up into smaller volumes and mount them where  
you see fit. Different volumes can have different options such as  
reserved space, quotas, compression, and others. In other words, it  
completely virtualizes your storage hardware in a flexible way and  
allows you to grow pools later if you add more disks..

2) ZFS includes a 256bit checksum (SHA-256) for each block. Data from  
a lost disk or corrupted block is reconstructed from this checksum  
automatically. This is there regardless of whether you use one disk  
for ZFS, mirrors or RAID-Z. There's a cool story where this  
checksumming really helped.[2]

3) It does snapshots on a per-volume basis. Making a snapshot is near  
instant and initially uses no additional space - it merely references  
the existing blocks in the volume. If/when a block changes its  
contents, the snapshot keeps a copy of the original for itself. You  
can also make unlimited snapshots of a volume. Where is this useful?  
For backups, of course. You can also split a snapshot off as its own  
volume (aka a clone).

There's more to ZFS than I feel like writing in this email, but you  
get the gist. Hit the two URLs I referenced.

Cyrus-specifically, the snapshots and compression features of ZFS are  
what Rob and I are eyeing. We're pondering moving our users' mail  
spool out of their AFS-based $HOME and to a separate Cyrus/MURDER  
system. We've always used AFS's snapshot feature to make a previous  
day backup of every user's home volume (and thus their email). We can  
still retain that feature with ZFS and it'll be both quicker and  
require less disk space. Initial testing shows that ZFS's compression  
can, on average, halve the amount of space a user's mail spool  
consumes, so there's an additional capacity win there. One more cool  
side effect of the compression is that your disks will be less busy  
since there are fewer blocks to read and write from the drives!

Oh, and if we need more space in the future, all we do is all more  
disks to the pool with a single command. BTW, ZFS is manipulated with  
just two commands: zpool and zfs. zpool is for maintaining your pool  
configurations and zfs is for manipulating the volumes within your  
pool(s).

[1] http://blogs.sun.com/roller/page/bonwick?entry=raid_z
[2] http://blogs.sun.com/roller/page/elowe?entry=zfs_saves_the_day_ta

/dale

--
Dale Ghent
UNIX Systems Specialist
UMBC - Office of Information Technology
ECS 201 - x51705