Last Updated February 02, 2007
A: Sun Cluster 3.2 will support ZFS via HAStoragePlus. It will only support fail-over for ZFS. October 30, 2006
Q: Have you played with ZFS in Sun Cluster 3.2 much? I'm curious how to make a ZFS file system global since you don't really "mount" a ZFS file system, and we're not supposed to use vfstab for ZFS in Sun Cluster 3.2. I can only see my pool (which is on a dual-hosted 3510 JBOD) from one node in the cluster. But I want all nodes to see it! A: Sun Cluster 3.2 does not support ZFS as a global file system. It only supports it as a fail-over file system. October 30, 2006
Q: I once experienced a neat feature in Veritas Volume Manager -- I did a new Solaris install (not upgrade) and installed the same or newer version of Veritas Volume Manager. Then we were able to have VVM inspect all the external disks and discover the volumes and recreate the device links automatically. Then we just reused the vfstab entries from the old Solaris install, and we were back up and in business. Does or will ZFS have the capability to (non-destructively) inspect disks and re-discover ZFS pools, etc.? VVM had a private partition on disks to hold its metadata -- does ZFS do anything like that? For example, if all ZFS storage on external JBOD and the host dies a bad death, we would like to attach the JBODs to a different host and bring up the ZFS pools, etc. (easily, with ZFS and Solaris OS doing as much work as possible). A: Yes, ZFS will do this. "zpool import" will list all the pools available to the system by searching all the attached disks for any pools. They can then be imported by name or by ID. November 6, 2006
Q: How long do you expect it will be before ZFS performance is on par with VxFS? A: I've not measured VxFS performance against ZFS. However, for general file system operations I would not expect ZFS to be wildly worse than VxFS now. Looking at the comparisons between UFS and ZFS, particularly where UFS is using directio, it is clear there is some way to go. But for more general file system operations the performance should be as good and often better than UFS. November 8, 2006
Q: I've started to play with ZFS for our web server but I discovered that I will have a big problem with the quota notion of ZFS compared to a traditional UNIX quota. I've tried to write something about this in my blog: http://blog.lgb.hu/post160. For quite a large redundant storage capacity (which can be extended online), with good FS snapshot-making ability, etc., ZFS would be perfect. However, on the storage itself I have several traditional UNIX daemons operating on discrete UNIX UIDs and need kernel-level quotas. There are more than 100,000 users, with heavy modifications done. I can't find a better solution than UNIX quotas (only an ioctl() is enough), which are not provided by ZFS. IMHO a UNIX environment should not break standard UNIX traditions, only extend them; this quota thing is an issue for me. A: Moving to a model where you have a file system per user is a change, but not one that daemons should have a problem with. Typically programs cope better with the concept of a file system being full rather than a user over quota, so apart from a daemon that is manipulating a user quota, there would be no requirement to change anything. It is true that currently the Solaris OS has difficulty managing very large numbers of file systems with an acceptable performance but that is being worked on and it should be possible to fix this. November 12, 2006
Q: There is a very strong case for file size quotas, not the least of which is mail spool files. We implement user-level quotas on our current VERITAS FS set aside for mail spool storage, with a 40 g per-user quota. This means that each user has a max of 40 gig of mail spool storage space at a file level. How do I implement this in ZFS, when the local mail delivery agent only knows how to deliver to files in a directory, not individual file systems or directories for each user? A: I'm not sure I agree that mail spool files is that strong a case. Yes, ZFS with respect to quotas for users changes things, and that may require changes to mail delivery programs, like sendmail. However, we already have people wanting to apply a single quota for a user that covers all their disk usage, both mail spool and home directory. People have in the past implemented this by making /var/mail a symbolic link that points into the same file system that contains the user's home directories. However, this does not scale beyond the case where all user directories can fit in a single file system. A much better solution would be to have the mail delivery program be able to deliver in to a file other than /var/mail/$LOGNAME. If you could configure this to be /tank/users/$LOGNAME/mail/in-box, and then have users' home directories in /tank/users/$LOGNAME/homedir, both could then live in the same file system /tank/users/$LOGNAME, or they could be separate file systems -- one for mail and one for the home directory, depending on the site's administrative choices. I have filed an RFE to have sendmail support this: 6488870 sendmail needs to support mail delivery to a directory (file system) per user. November 16, 2006
Q: What is the largest file system size currently supported for ZFS? A: The limit is practically defined by the size of the storage you can attach to a system. The limits are:
December 07, 2006
Q: So user quotas will not be implemented... Well, that brings us to a position of an unusable file system for user purposes, because how can I export 50,000 file systems via NFS, one for each user? A: It is a bit early to jump to that conclusion. There is work going on to make issues around the thousands of file systems go away. When NFSv4 clients can cross server mount points without needing the automounter and a server with many thousands of file systems can boot in a reasonable time, the need for user quotas disappears. ** Information is outdated. Please see Note. December 12, 2006
Q: With Solaris 10 (11/06) out, what has changed with regards to ZFS from its inital release? A: The following patches will get you to the same level of ZFS support as there is in the Solaris 10 (11/06) release. Please note there may be later versions of patches available: Solaris 10 Update 3 (11/06) Patches SPARC Patches
These patches deliver the following new features and bug fixes: ZFS Features/Projects
ZFS Bug Fixes/RFEs
January 26, 2007
Q: We use EMC SAN storage, which internally handles mirroring and RAID, and we are looking at using ZFS on these. Is there a best practice for using ZFS on SAN storage? A: Ideally you should present simple striped LUNs to ZFS and let ZFS handle the redundancy so that it can spot any data errors on one path and recover from the other. Even if you cannot do this and the EMC storage is doing the redundancy, then ZFS will still check the data. However, since there is no second copy available to ZFS, if there were a data integrity issue, it would return an error to the application. Currently such a configuration should be avoided because if you have a write to the device and you lose all the paths to the device, the system will panic. Obviously if you have multiple paths, which would be normal, the probability of this is small. Also other file systems don't cope much better if at all. Nonetheless, ZFS is being worked on so it would be able to recover from such a failure more gracefully than with a panic. February 02, 2007
Question file was not found. |
 
BigAdmin SubscriptionsBigAdmin Areas
BigAdmin Sun Center
BigAdmin Topics | |||||
| |
