Fast Track to Solaris 10 Adoption: ZFS Technology

Please click on a question below or download a pdf version.

1. What types of QOS features does ZFS have in terms of data availability? Can it be configured with a minimum service time for requests?
2. If an existing Solaris 8 OS server is upgraded to the Solaris 10 OS using the install upgrade process, will it be possible to include a migration to ZFS at that time?
3. Will I be able to upgrade a server running the Solaris 9 OS to the Solaris 10 OS with ZFS? If so, what are some of the caveats?
4. How do I configure swap for ZFS?
5. How does ZFS set up? How does it choose RAID type for instance?
6. Will it be possible for me to easily migrate my VxVM volumes to ZFS? If so, how?
7. If you configure ZFS to a single pool, would you be able to use part of that pool as additional swap space when the need arises?
8. Can ZFS be set up to prefer certain devices over others in its storage pool? If so, is this set at the pool or filesystem level?
9. In reading the available information on ZFS and your visions of it, I get a little confused about how I should be planning my storage systems if I plan to use ZFS. If you were planning about 6 TB of storage, what would be most in line with the ZFS thinking and why?
10. What are the tuning parameters you're offering initially? Anything like direct I/O, concurrent I/O, read/write release behind, etc. mount options?
11. Is storage grouped into a small number of storage pools with intuitive names? Is each storage pool represents QOS type?
12. What kind of code is managing the "storage pool"? Would it become a single point of failure? In traditional LVM, one volume failure doesn't affect others. What happens when "storage pool manager" has a problem?
13. What conversion tools if any will be needed?

Q: What types of QOS features does ZFS have in terms of data availability? Can it be configured with a minimum service time for requests?

A: We have the ability to inherit a priority for I/Os, which are then taken into account when doing our disks I/O scheduling. This scheduling takes into account both deadlines and priority with the goal of minimizing overall system latency and maximizing throughput to the disks. We are investigating what kind of interface to provide to the administrator to further control the priority given to I/O requests from an application/user/zone.

Back to top

Q: If an existing Solaris 8 OS server is upgraded to the Solaris 10 OS using the install upgrade process, will it be possible to include a migration to ZFS at that time?

A: ZFS can be added to the system as part of the upgrade, but existing UFS filesystems will remain UFS. Data migration tools are planned, but will not be available when ZFS first ships.

Back to top

Q: Will I be able to upgrade a server running the Solaris 9 OS to the Solaris 10 OS with ZFS? If so, what are some of the caveats?

A: To migrate your storage from UFS to ZFS, you will need to do a backup (using a POSIX-compatible backup utility) from the UFS filesystem and then restore onto ZFS.

Back to top

Q: How do I configure swap for ZFS?

A: We have not published those processes yet; this will be described in the ZFS documentation when we make that feature available in a Solaris 10 OS build.

Back to top

Q: How does ZFS set up? How does it choose RAID type for instance?

A: The administrator specifies what type of RAID they want. We will provide control such that you can either have detailed control over which disks are replicated with which, or you can just have a "bag of disks," and we will do our best to match disks up based on the administrator's request.

Back to top

Q: Will it be possible for me to easily migrate my VxVM volumes to ZFS? If so, how?

A: Right now, it's dump and restore; migration tools are on the project list, but won't be there until after ZFS first ships.

Back to top

Q: If you configure ZFS to a single pool, would you be able to use part of that pool as additional swap space when the need arises?

A: Yes

Back to top

Q: Can ZFS be set up to prefer certain devices over others in its storage pool? If so, is this set at the pool or filesystem level?

A: The filesystem/pool interface follows a malloc/free paradigm. Any such control would be at the pool level. The device preference is currently controlled by our adaptive algorithms. We are exploring ways to let the administrator give us "hints" to tune these algorithms, but we have not finalized what will be in the final version.

Back to top

Q: In reading the available information on ZFS and your visions of it, I get a little confused about how I should be planning my storage systems if I plan to use ZFS. If you were planning about 6 TB of storage, what would be most in line with the ZFS thinking and why?

A: In most situations, we configure all storage into a single pool. This makes it simplest for the administrator and allows all filesystems on the system to share both space and bandwidth.

Back to top

Q: What are the tuning parameters you're offering initially? Anything like direct I/O, concurrent I/O, read/write release behind, etc. mount options?

A: Our goal is to make such tuning parameters unnecessary — ZFS should be able to obtain optimal performance with a minimum of "wacky knobs."

Back to top

Q: Is storage grouped into a small number of storage pools with intuitive names? Is each storage pool represents QOS type?

A: ZFS is quite flexible, and you can configure your system that way if you wish. We can support any number of storage pools, and if you want to configure each pool according to QOS, we support such a configuration. And each pool (as well as each filesystem) has a name that you pick, so they are as intuitive as you wish to make them.

Back to top

Q: What kind of code is managing the "storage pool"? Would it become a single point of failure? In traditional LVM, one volume failure doesn't affect others. What happens when "storage pool manager" has a problem?

A: It all depends on how you configure your storage pool. If you configure it such that you have redundant copies of your data, we can recover in the case of a failure. As long as at least one copy of your data is online, we can service any requests to it. If you're referring to a software failure (bug, panic, etc.), then that may cause your system to reboot, just like a failure in traditional LVM kernel code.

Back to top

Q: What conversion tools if any will be needed?

A: If you want to convert your data from an existing filesystem to ZFS, you will have to copy the data from your existing filesystem over to ZFS. At some future date, we plan on providing a mechanism to migrate your data to ZFS online.

Back to top

Next >