Honestly, i think ZFS is the best choice if you care about your data. Also, being that ZFS is a fully software solution you don't need to spend a ton of money on raid cards. It gets speeds you wouldn't believe. The only downside to ZFS is that you have to plan your system out smarter than most other raid solutions but the advantages far outweigh the disadvantages. Let me list a few features:
ZFS is a pooled storage system. you have a zpool which holds your zfs filesystems (you can have multiple filesystems)
All filesystems use space as they need it from the pool. you can put quotas and reservations on the filesystems if you like but you don't have to format them or preallocate space. It works in a similar manor as virtual memory works for memory.
ZFS is a copy on write system. It always writes new data instead of the write/modify/read system that most other filesystems use. Because of this, you always have a consistent copy of data online. Writes either happen or they don't. This eliminates the raid5 write hole that other systems have and fully removes the need for tools like FSCK.
Because of it's copy on write nature, you can take "moment in time " snapshots and they happen almost instantly. A snapshot is a full read-only copy of a filesystem at the time it was taken. It takes up almost no space at first, and only grows in size based on what changes, and if you delete something from the live data set you can recover it from the snapshot.
You can clone snapshots making a second filesystem which is read/write. It takes up no additional space except for any new data you add to it.
A real world example of how i used this was, i had a filesystem with all my music in it, i cloned it for my girlfriend who deleted what she didn't like and added in what she did
ZFS can do on the fly compression. It will compress the data to disk, read the compressed data and it happens without the need to decompress or recompress. It's really great for lots of data sets because with todays modern cpus you have plenty of extra CPU cycles, while the disk i/o is the real bottleneck, so in a lot of these situations, you actually GAIN speed.
ZFS now has dedup which allows you to only keep one copy of data on file and all like data has a pointer to that data. This is awesome for virtual machine images and such. It works on the block level, not the file level, so a lot of times dedup can save a lot of space. This depends on your data though, like compression. Probably requires a more hard core machine though.
The best thing about ZFS is that all data and metadata is checksummed. each checksum points to the next, and the entire thing forms a self validating merkle tree. In english, this means ZFS can detect, and fix corrupted data that other filesystems miss. I've seen this in action and it's amazing. This saved an entire pool for me recently due to a failing sata card which was sending bad data. But this also protects against disks that have blocks fail slowly over time. As long as you use redundancy, this can fix those errors and it works better than ANY other filesystem available today. This is the number 1 reason i'd never use anything else. If you care about your data, you will use ZFS. While raid and zfs aren't a substitute for good backups, it's a helllllll of a lot more safe.
ZFS is a software only raid system (you can use hardware raid with it but there is no real advantage) ZFS does what other software raid systems can't really do. It basically uses your ram and cpu to form a giant hardware raid system. It makes intelligent choices about disk i/o. A perfect example is, i have a system with 20 hard drives that gets 600-700 MB/s reads and 550-600 MB/s writes. While this is FAR more than gigabit, it is very nice (though i have mine trunked with 2 lines right now so i get about 200 MB's)
ZFS makes it really easy to share filesystems as windows filesystems or NFS filesystems if you are using it on Solaris (if you use FreeBSD then you can still do this but it will require samba or nfs to be set up)
on solaris you can share a filesystem with windows just like this:
Code:
zfs set sharesmb=on tank/naswhere the filesystem is named tank/nas Now that filesystem is available for windows and you can set it as a network drive. Simple.
You can also create zvols on a zfs pool. A zvol is a virutal block device (a virtual hard drive) You can use these zvols as iscsi hard drives or to install virtual machine images on. It's quite cool. I use one for my mac's time machine volume. These gain all the other features of ZFS liek snapshots, compression and checksumming.
To be fair, ZFS DOES have some disadvantages. One. It only works on Solaris 10 OpenSolaris and FreeBSD (though the last freenas has it i think)
OpenSolaris is probably the best to use because it has the newest feautes. FreeBSD works on more hardware though, but if you are building a system this is a non-issue. If you are trying to use old hardware, it may be more of an issue.
Another limitation is that you should REALLY use a 64 bit processor. These days this isn't as much of an issue, but it is still a problem. while it WILL work on 32 bit cpus, its not a great idea, because 32 bit has very limited amounts of addressable kernel ram
ZFS loves ram. While this isn't as much of a con, it can be to some people. I wouldn't use less than 4gb ram (though i know pleny who use 2gb, and technically 1 gb is the minimum) My personal system uses 8 gb. Also, it's best to use ECC ram if possible. Thsi saves against bad memory causing errors and keeps your checksums correct. What good is self healing if the memory can't be trusted.
For the next one, i need to explain a little about how a zpool works. Like i said earlier, you have a pool which holds yoru filesystems. This pool is made of what is called vdevs. a zpool can be made of one or many vdevs. a vdev is any of the following:
a hard drive
a partition on a hard drive
a raidz(1,2,3) group
a mirrored group (2 way 3 way, n way)
a file (though this is mainly just used for testing)
ZFS will stripe data across VDEVS in a dynamic fassion. now, it's best to use redundant vdevs for what we are talking about. This means mirrors or raidz (raidz1 is similar to raid5 (well it's a ton more advanced, but for your understanding, thinking of it as raid5 is file) raidz2 is like raid6, raidz3 is...well, tripple parity)
Currently, there is no way to "grow" a raidz(123) vdev. If you have a vdev with 5 drives, you can't add a 6th, though you CAN always add more vdevs. This is what i meant about having to plan your storage better. If you know you can't buy 5 and 6 drives at a time, and can only afford to buy 2 at a time, then you should use mirrors. (mirrors are faster than hell anyways, though honestly, i'd go with raidz2)
I personally like raidz2 because it gives you really good space, damn good speed and nice fault tolerance without killing you in space.
raidz1 is ok too...if you can only get 4-5 drives at a time, then raidz1 may be the way to go. If a single drive fails, you are ok with raidz1
My pool looks like this:
Code:
pool: tank
state: ONLINE
scrub: none requested
config:
NAME STATE READ WRITE CKSUM
tank ONLINE 0 0 0
raidz2-0 ONLINE 0 0 0
c5t0d0 ONLINE 0 0 0
c4t1d0 ONLINE 0 0 0
c6t1d0 ONLINE 0 0 0
c5t1d0 ONLINE 0 0 0
c6t3d0 ONLINE 0 0 0
c4t0d0 ONLINE 0 0 0
c5t3d0 ONLINE 0 0 0
raidz2-1 ONLINE 0 0 0
c6t0d0 ONLINE 0 0 0
c5t5d0 ONLINE 0 0 0
c6t2d0 ONLINE 0 0 0
c4t3d0 ONLINE 0 0 0
c5t2d0 ONLINE 0 0 0
c6t4d0 ONLINE 0 0 0
c4t2d0 ONLINE 0 0 0
raidz2-2 ONLINE 0 0 0
c5t4d0 ONLINE 0 0 0
c6t5d0 ONLINE 0 0 0
c5t7d0 ONLINE 0 0 0
c4t5d0 ONLINE 0 0 0
c6t6d0 ONLINE 0 0 0
c5t6d0 ONLINE 0 0 0
cache
c8t5d0 ONLINE 0 0 0
spares
c4t7d0 AVAIL
If you have any questions, ask and i'll explain. Trust me though, ZFS crushes everything else.
