Last Updated February 02, 2007
flarcreate?
A:
You can create a flash archive from a system that is using ZFS, as a flash archive is just an archive of the data on the system, using ** Information is outdated. Please see Note. October 9, 2006
Q: Still on user quotas. So there is no loss of performance, having 10000+ ZFS file systems for each user, over the old system of having one file system and user quotas? A: Yes. ZFS is designed to scale to many thousands of file systems with the intention that each administrative entity, user, group, project, etc. has its own file system. UFS user quotas don't really scale well on file systems with large numbers of inodes due to quotacheck having to scan the whole file system at mount time, massively increasing the boot time. ** Information is outdated. Please see Note. October 12, 2006
Q: We are implementing a test ZFS pool in two Sun StorEdge MultiPack devices. During the file system migration (zfs send/receive) from the test ZFS pool to a production ZFS pool (F-CAL to Sun StorEdge A5000 system), the server crashes due to CPU panic. Both pools are running RAID-Z. Is there another migration option? A: Clearly that the system crashes is a bug. If you have a crash dump, can you let me know how I can access it. Even if you can only send the panic messages and the stack trace from the messages file, that would be a start.
Until this bug is fixed you will have to revert to the more traditional methods of copying data from place to place. October 12, 2006
Q: Will a ZFS pool automatically expand, if the size of a LUN or other hardware RAID device beneath it is grown? (Does it have the functionality of the UFS growfs command, or is there a ZFS equivalent?)A: Almost. Once the label on the disk is updated to reflect the new size, the pool will automatically find the new size when it is next imported or where you have redundancy via mirroring or RAID-Z when the devices are brought online. There is work going on to improve this. The following bugs describe some of that work: 4967547 format doesn't notice when EFI labeled devices change size 6475340 when lun expands, zfs needs to write new label October 12, 2006
Q: When is it expected that ZFS will support the ability to shrink a pool (remove a disk from a pool without replacing it)? A: This feature is expected to be available sometime in 2007 (in Solaris Express; support for the Solaris 10 OS will follow). ** Information is outdated. Please see Note. October 12, 2006
Q: Is ZFS cluster aware, for example, can we use it in place of VERITAS file systems? A: Yes. Sun Cluster 3.2 will support a cluster-failover solution for ZFS. October 19, 2006
Q: There are many types of file systems, for example, UFS, Ext3, VxFS, QFS, and SAM-FS. Each has its methods of allocating data blocks. I would like information on how ZFS allocates data blocks and on its data structure. This is from a performance perspective: How do we position ZFS performance compared with VxFS, UFS, Ext3, QFS, and SAM-FS? A: ZFS is different from the other file systems in that it operates a copy on write policy. So each time a block in a file is written, it is written to a different location in the pool. This also allows writes to be clustered so that they can be sequential and therefore faster. For details of the current implementations, you should visit the opensolaris.org ZFS Community. October 30, 2006
Q: Will ZFS (or more accurately, storage pools) support dynamic device removal? I want to remove a 72gb disk from an unmirrored storage pool that currently has three 72gb disks. There is enough space to remove one drive and still hold all the data. A: Yes, in the future, removing vdevs from pools will be supported. It is expected to be available sometime in 2007 (in Solaris Express; support for the Solaris 10 OS will follow). ** Information is outdated. Please see Note. October 30, 2006
Q: How long will it be until you can mirror boot disks with ZFS the way you can with SVM? A: This is really how long before you can have your root disk on ZFS. 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. ZFS root file system support is aimed for Solaris 10 OS Update 4. ** Information is outdated. Please see Note. October 30, 2006
Q: We tried implementing ZFS with 30,000 users with quotas per user on a SF V440 connected to a SE6920. Rebooting the server took approximately four hours to unshare/unmount each user's file system. At this point the old UFS implementation with quotas seems much faster. Are we missing something here? A: This is a known issue -- well, two known issues, actually. The first is that the initial mounts of 30,000 file systems takes a long time, 37 minutes on a Thumper in the lab here. This is due to this bug: 6386929 initial "zfs list" is slow Most of the time is spent waiting for I/O, and we expect to be able to improve this to an acceptable number. The second issue is that it takes a very long time to share 30,000 file systems. This is being worked on as part of the "shreamgr" project, although it won't be completely resolved by the first drop of the code into OpenSolaris. October 30, 2006
Q: I have some problems with mounting ZFS: I've created a zfspool and some file systems without any problems. But at the next boot no ZFS file systems are mounted! I used zfs mount -a. The mountpoints are all custom -- no "legacy" or "none" mounts. A: Is the pool being imported? What does zpool list return? Are there any core files around? If so, then run the file on it, and if it is from zpool, run pstack on it. It could be this bug: 6460398 - zpool import cores on zfs_prop_get. October 30, 2006
Q: On non-production servers, I frequently have ufsdump-based scripts for backups. Will Sun have a "zfsdump/zfsrestore" equivalent, or will ufsdump be patched to work on ZFS partitions? I have always liked the manual intervention with ufsrestore for selecting certain files for restore. If there will be no equivalent to ufsdump/ufsrestore, would the recommended solution be to use snaps on ZFS and then tar/pax and so on? A: The closest equivalent commands to ufsdump and ufsrestore would be zfs send and zfs receive. However, as you rightly point out, the recommendation would be to use snapshots and then tar, cpio or pax to back up the snapshots. If you have the disk space to spare then the ideal is just to keep the snapshots online. October 30, 2006
Q: Is ZFS compatible with SAM-FS/QFS file systems, and if so, how much trouble is it to convert existing file systems to ZFS with SAM-FS? A: There is no integration of SAM-FS and ZFS. You could configure SAM-FS to write archive copies to ZFS in the same way as you can configure it to write archive copies to any UNIX file system. See the Disk Archiving section of the Sun StorEdge SAM-FS Storage and Archive Management Guide. October 30, 2006
Q: Suppose you have five disks in a RAID-Z configuration and one of them fails. According to the manual, you have a degraded RAID-Z set, and redundancy is lost. Wouldn't it be nice if the system could regenerate its redundancy and fall back to a smaller size? Since with ZFS the file system is decoupled from the storage elements (disks), if there is enough free space, the pool can scale down to a RAID-Z with four disks. Are there any plans to implement a feature like that? A: It may be a nice feature to have, but I'm not sure it would be all that useful in most cases unless you wished to shrink the pool anyway. You may want to wander over to the OpenSolaris ZFS community and ask what they think. To my knowledge, there are currently no plans to implement such a feature. October 30, 2006
Q: Our databases, both Oracle and Sybase, use raw partitions. I would love to move them to ZFS. Can this be done? If so, have there been any performance tests done on database ZFS vs. raw partitions with async I/O? A: ZFS provides zvols for applications that want access to a device rather than a file system. See zfs and the -V option. Sun's Performance engineering group (PAE) has done some comparisons of database performance on UFS and ZFS -- you can read Neelakanth Nadgir's blog - Databases and ZFS. October 30, 2006
|
 BigAdmin SubscriptionsBigAdmin Areas
BigAdmin Sun Center
BigAdmin Topics | |||||
| |
