** Note: Some information in these answers is outdated. For up-to-date information about ZFS, see the ZFS FAQ.

 
 
 
Q: What is ZFS?

A: ZFS is an innovative ground-up redesign of the traditional UNIX file system. It has end to end checksumming to protect against data corruption and a simple administrative model.

ZFS eliminates the use of volume managers instead using pooled storage where many file system can share a pool.

September 26, 2006 Back to top

---

Q: On what versions of Solaris is ZFS included?

A: ZFS is available in Solaris 10 06/06 and later releases. It is also available on Solaris Express and OpenSolaris.

September 26, 2006 Back to top

---

Q: Can I add ZFS to older versions?

A: Yes you can to a Solaris 10 install. If you apply patches 122640-05 for SPARC or 122641-06 for x86 to your Solaris 10 system you will have support for ZFS added to your system. However the preferred method is to upgrade to Solaris 10 06/06. This will save you having to worry about the many patch dependencies.

September 26, 2006 Back to top

---

Q: What incompatibilities do I have to worry about?

A: ZFS is a posix compliant file system. The only changes users may notice from UFS are quotas and Access Control lists.

ZFS no longer has the concept of a per user quota. Instead the quota is applied to a file system to limit the file system's size. Therefore to have quotas for users you have to implement a file system per user. Thanks to the pooled storage and the improved file system management creating file systems can now be done with a single, simple command.

Access Control Lists (ACLs) are handled differently on ZFS. Instead of using the getfacl(1) and setfacl(1) commands to read and manipulate ACLs you now use the ls(1) (ls -v) command and the chown(1) commands to do so. The new ACLs are much finer grained than the ones available on UFS.

September 26, 2006 Back to top

---

Q: I am implementing network file storage based on ZFS. Clients (Solaris OS, Linux, and MacOSX) will use NFS to access Apple Xraid, which is a hardware platform for storage.

I really like the idea of file system separation (per project, per user, and so on). But I will lose browsability of the NFS-exported file system, say /export/project, if the file system contains a child file system that is NFS-exported as well. I can survive using automount of users' home directories, but the main file storage tree is useless without the ability to browse if via NFS.

I would like to use file system separation (/export/project, /export/project/A, /export/project/B...) and share /export/project via NFS so that users can see and browse projectA and projectB directories while accessing them via NFS.

A: While ZFS exposes this issue in a way that was not previously seen, the issue is not limited to ZFS but applies to any system with a large number of file systems.

There are currently three possible solutions:

The best solution is to have an automount map for that file system, which you update when you create a new file system. This will allow you to browse the file system and have it dynamically updated when you create new file systems.

If you don't expect to create a lot of new file systems then you can use the /net automount point to access the new file systems. However, if you have a host called foo with a pool tank mounted on /tank and you change the directory on a client to /net/foo/tank, you will be able to see all the file systems that are shared below tank. But if another new file system is created under tank, it will not be visible on the client until the client unmounts /net/foo and can remount it. So for an environment where you are not dynamically creating file systems often, this could be used.

If you require dynamic updates and will not use a separate static map, then the best solution would be an executable automount map. However, the drawback with executable maps is that they cannot be browsed.

September 29, 2006 Back to top

---

Q: I'm using SVM on my root partition. Is there a UFS-to-ZFS conversion tool?

A: No, there is no tool to convert from SVM and UFS to ZFS. Currently ZFS support for root file systems is only in Solaris Express and is limited.

To move existing data to ZFS requires some extra space to create a zpool and then you have to copy the data. Once you free up devices from SVM they can then be added to the zpool, allowing more data to move over.

September 29, 2006 Back to top

---

Q: I have installed SXCR build 46. I have a spare empty disk. Can I get ZFS running on it without breaking my existing installation?

A: Yes. If you have spare disks or partitions they can be used for zpools, which in turn will hold ZFS file systems without impacting your existing system setup. (Just like you could put a UFS file system on them if you wished.)

September 29, 2006 Back to top

---

Q: Can ZFS be used on the root disk?

A: Not yet in the Solaris 10 OS.

There is support for having an x86 system's root disk in OpenSolaris (http://www.opensolaris.org), however it is still under development, so things like upgrade and live upgrade are not there yet.

** Information is outdated. Please see Note.

October 4, 2006 Back to top

---

Q: Given two servers and a single LUN from a 6130 array, currently I only know of two ways to correctly share the LUN, either using QFS or NFS. Can ZFS be utilized in a similar way given its data integrity features, or are NFS or QFS still required to achieve this?

A: If you wish to have simultaneous access to the same ZFS file systems on more than one system then NFS is still required.

October 4, 2006 Back to top

---

Q: Will per-user quotas be available in future versions of ZFS?

A: There are no plans to support user quotas as we have in UFS in the future. The model is that each user is given their own file system; now that file systems are so cheap and easy to create, this is not the problem it was with UFS. A quota can then be set on that file system restricting that user to that space.

So the old model of having a large file system with lots of directories and user quotas becomes a large zpool and then lots of file systems, one per user, with a quota for each user file system.

Using the hierarchical nature of the ZFS file systems quota, you can even implement group quotas by having a file system per group and then the users in that group having file systems in that group file system:

File System             Quota
tank
tank/admin        	20G   
tank/admin/user1        10G
tank/admin/user2        10G
tank/admin/user3        10G
tank/eng                50G
tank/eng/user4          20G
tank/eng/user5          20G

In this case even though each admin has a quota of 10G the total for the group is only 20G and when you report the usage using df(1) it reports what is currently available to the user. So if user1 and user2 both use 8G, when you run df(1M), it will show that user3 can only use 4G while user1 and user2 only have 2G available:

File System             Quota    Used    Available
tank
tank/admin       	20G       16G      4G
tank/admin/user1        10G        8G      2G
tank/admin/user2        10G        8G      2G
tank/admin/user3        10G        0G      4G

** Information is outdated. Please see Note.

October 4, 2006 Back to top

---

Q: Is ZFS supported with Oracle, and if so, what versions? Is it also compatible with Oracle's ASM?

A: This is really a question for Oracle. ZFS is a POSIX file system, so Oracle will run on top of it and work, and there should be no reason for Oracle not to support it.

Oracle ASM is a replacement for a volume manager and file system replacement to provide the flexibility of a file system. So it is not really clear why you would want to use it with ZFS.

October 4, 2006 Back to top

---

Q: Will there be a zfsdump similar to ufsdump? There is zfs send and zfs receive but I believe we need a bare metal restore, especially when the root file systems are placed under ZFS.

A: There are no plans for a zfsdump similar to ufsdump.

You can use zfs send to generate a backup of a file system. However, you have to be aware that the output of zfs send could change in future releases, so it's possible that you may not be able to restore such a backup if you upgrade ZFS in the future.

However you can use cpio(1) or pax(1) to generate a backup of a root file system to back up and restore data into ZFS file systems, including root file systems. Indeed the current method of creating a ZFS root file system involves using cpio(1) to copy a root file system into a ZFS file system.

October 9, 2006 Back to top

---

Q: As with anything new, performance questions are always near the top of the list. I have found a number of Sun blogs discussing ZFS performance characteristics, but it is rather difficult to find official performance benchmarks when comparing ZFS with UFS, VxFS, raw, and so on, in the context of various workloads such as Oracle databases or NFS servers. Do these exist?

A: Currently the only comparisons we have are the ones on the blogs that you have found.

In particular, Roch Bourbonnais's blog has comparisons for ZFS, and Neelakanth Nadgir's blog has recommendations for improving the current ZFS performance.

** Information is outdated. Please see Note.

October 9, 2006 Back to top

---

Q: Besides the root file system, what other file systems should ZFS not be used on in the current release of Solaris (06/06)?

A: You should avoid using ZFS for any file systems that are involved in an upgrade. So, for example, having /usr, /opt, or /var on ZFS should be avoided.

However, having /opt/sfw, /var/mail, or /var/spool/mqueue on ZFS is fine.

** Information is outdated. Please see Note.

October 9, 2006 Back to top

---

Q: What new features are likely to be added to ZFS in the next release of Solaris?

A: Here are the new features that have been ported to the next release of Solaris 10:

• Support for hot spare disks in a pool
• The ability to promote a clone of a snapshot to be the parent of that snapshot
• Recursive snapshots
• Double-parity RAID-Z

Plus many bug fixes. A complete list can be found on the opensolaris mail archive: http://mail.opensolaris.org/pipermail/zfs-discuss/2006-July/004050.html.

** Information is outdated. Please see Note.

October 9, 2006 Back to top

---