Comments on: ZFS On Mac: Now All-But-Official

By: Dave Mount

Dave Mount — Thu, 17 May 2007 01:41:06 +0000

Robin, I wanted to comment off the zdnet posting as it tends to get zealot crazy, that I appreciate your article. Thanks!
Also, as someone who earns a living from selling Apple systems and OS X-native apps to professionals (medical and dental), it is important to me to be able to provide them with information that will be an incentive to have them invest in Apple technology. With ZFS implemented in OS X, the braininess of Sun engineers will be of practical benefit to my clients whose entrustment of *their* clients’ medical information, is of utmost importance. Just another buttress in the system that will convince users that the Mac platform design is all over what Jobs’ “user experience” is all about!

By: Todd

Todd — Fri, 02 Feb 2007 02:50:53 +0000

As of January 2007 Solaris zfs backups are rudementary and not why I used zfs. In most places day to day restores are file recovery operations. Recovering an entire filesystem should only be necessary following a massive failure. All zfs restores depend on restoring a full filesystem. There are ways to make a zfs backup system work, but they are not beautiful like raidz2, three way mirrors and instant filesystems.

Basically you snapshot the filesystem and send the snapshot data to a tape or a file or to another zfs command. There don’t seem to be any error messages. What I mean is that you don’t really know if your backup is any good. How’s that from the people who checksum the data on every write and protect it with double RAID and hot spares.

Let’s assume the backup worked. If you want to restore a file you need to pull the entire filesystem off the tape and convert it into a file system:
zfs receive tape or a file or a pipe”
The result of sending a zfs snapshot to a tape or file is a single large file. There is no way to extract a particular file or get a list of the contents. You can send it through a pipe to “zfs receive somenewzfs”. It creates an entire copy of the filesystem, which is great when you have the space - night after night.

By: Robin Harris

Robin Harris — Mon, 18 Dec 2006 22:28:51 +0000

I suspect it will become the default sooner than 10.6. Home users actually put a lot less stress on a file system than servers do. So if they’ve got the uitlities built for homies to manage it, I’d guess 10.5. 4.. The Solaris version has been out for a year now, so it is getting solid.

I don’t know what they are going to do with the Xserve RAID.. FC is great for distance and speed, but I think a lot of folks would be happy with SAS and saving another several hundred bucks. Yet if you run RAID 0, you have access to all the drives, and you need the controller to MUX the FC-to-ATA traffic anyway. Of course, they could run FC-AL, but I would have to believe that would be a support nightmare for them, unless you limited the number odrives to a real small number.

Given that Leopard has native iSCSI support, another interesting possibility emerges for the low-end. How about, and you heard it here first! - an iServe RAID. ISCSI RAID box on Gig E. Same packaging, different controllers. $100 Gig E switch instead of the $1400 FC switch. Slower, but a lot of folks won’t care.

Robin

By: Wes Felter

Wes Felter — Mon, 18 Dec 2006 19:57:24 +0000

Now will Apple replace Xserve RAID with Xserve JBOD?

By: hmurchison

hmurchison — Mon, 18 Dec 2006 16:48:17 +0000

I was wondering when you’d write a piece on the new ZFS info. It looks like a nice start and perhaps by 10.6 ZFS will be the preferred fs for the entire OS. I would not have a problem with that. We don’t need RAID controllers when in 2006 the avg Mac will have 4 cores.

ZFS is important as we begin to move towards Terabyte drives. Keeping that massive amount of data incorrupted is important.