Zfs for mainstream linux finally!

[darshin]

Hi Everyone,

Many thanks for your contribution, support and patience. We had an overwhelming response to the closed beta for ZFS port on Linux. The reviews of the closed beta have been very promising and a big thank you to all the people / companies who have been involved in testing the closed beta and submitting their feedback and comments.

We had mentioned earlier that once we have enough reviews about the closed beta, we would be making the ZFS port on Linux beta an open beta for everyone. So here we go folks! THE BETA IS PUBLIC AND OPEN FOR EVERYONE NOW! Please login to http://zfs.kqinfotech.com and register to download. No more waiting for approvals from our team.

However, I would like to get a regular feedback from you all, once you start testing the beta. Your feedback and bug reports would be very important as we now inch closer to the GA Release which is scheduled for December. Please feel free to contact me at *[email protected]

Cheers,
Darshin

[darshin]

What are the read and write speeds in your raidz2? tell the number of disks and models too, please.

The Beta is available to test. Logon to zfs.kqinfotech.com and register for it.

[marakaid]

I've been running a ZFS-FUSE RAIDZ2 for some time on my Ubuntu server, and it's been rock-solid.

What are the read and write speeds in your raidz2? tell the number of disks and models too, please.

[kernelOfTruth]

True, but the effort to get either a ZFS-FUSE or native ZFS pool running on Linux is similar to that required to run BTRFS currently. Sure, FUSE is slower, but reliability is more important to me than performance for the kind of data I'm using ZFS for.

not really: for me it has the same speed like btrfs or ext4, you only need to "donate" some big amount RAM (e.g. 1-2 GiB) of it to really see nice results (you could also do it with 128 MB given to it but under my impression it's better if more is in cache) - it's highly threaded so the more cores or threads your CPU can handle the better

besides that:

- there's fuse splice-support in newer kernels now (afaik zfs-fuse doesn't support it yet - so people might want to help to get that to work)
- a parallel (pcrypt) version of the crypto-stuff is in the kernel for those who aren't fortunate to own an CPU with AES hardware-acceleration
- zfs-fuse has gained much speed since 0.6.9
- ...

[tokyovigilante]

Looks like there is a zfs-fuse package, maybe I'll try it sometime.

zpool create <poolname> /dev/<disk>.

Admin is pretty straightforward. You can add vdevs (ie add disks) to a pool once created (eg for JBOD), but raid configurations can't be changed once the pool exists, although drives in a RAIDZ can be replaced with larger ones.

[movieman]

Looks like there is a zfs-fuse package, maybe I'll try it sometime.

[movieman]

True, but the effort to get either a ZFS-FUSE or native ZFS pool running on Linux is similar to that required to run BTRFS currently.

For btrfs, in Ubuntu I just type 'mkfs.btrfs /dev/whatever' and mount the partition.

I've no idea whether there's even any kind of zfs support I could install, let alone how to set it up?

[tokyovigilante]

I haven't really noticed a bias myself, but there's just not much point getting enthusiastic about a filesystem that isn't included in the most common Unix-like OS on the planet.

I suspect most of us would run ZFS if it was an option, but it's not. Sure, there's the FUSE port, but from what I've read the performance sucks and it's probably nowhere near as robust as a native kernel implementation.

Personally I was going to build an OpenSolaris server at home just because of ZFS, but now that Oracle's taken over Sun I don't want to get stuck with an operating system that they suddenly stop supporting.

True, but the effort to get either a ZFS-FUSE or native ZFS pool running on Linux is similar to that required to run BTRFS currently. Sure, FUSE is slower, but reliability is more important to me than performance for the kind of data I'm using ZFS for.

[movieman]

There seems to be a real BTRFS bias among the Phoronix writers. I guess this is a Linux forum after all, but ZFS is currently a superior filesystem.

I haven't really noticed a bias myself, but there's just not much point getting enthusiastic about a filesystem that isn't included in the most common Unix-like OS on the planet.

I suspect most of us would run ZFS if it was an option, but it's not. Sure, there's the FUSE port, but from what I've read the performance sucks and it's probably nowhere near as robust as a native kernel implementation.

Personally I was going to build an OpenSolaris server at home just because of ZFS, but now that Oracle's taken over Sun I don't want to get stuck with an operating system that they suddenly stop supporting.

[tokyovigilante]

There seems to be a real BTRFS bias among the Phoronix writers. I guess this is a Linux forum after all, but ZFS is currently a superior filesystem. The RAID support is mature, and it has been in production several years.

I've been running a ZFS-FUSE RAIDZ2 for some time on my Ubuntu server, and it's been rock-solid.

BTRFS is still in early beta, is missing object-level RAID, and seems to have very poor userspace utilities. ZFS has a 5 year head start. BTRFS may certainly catch up, and the ongoing development of ZFS is uncertain given the demise of OpenSolaris, but currently ZFS is a better choice for reliable redundant storage. Improved raw performance on a few benchmarks means little without safety.