In 2008, I had to explain what ZFS was and why it’s different to the existing volume/filesystem model. By the end of 2009, IT Pros will all be aware of it, what it does and will likely have at least a little of it in their production infrastructure. Sun has already started that ball rolling.

Heck, with full-blown support of ZFS likely to be in OSX 10.6, “Snow Leopard”, it’ll even make in-roads to the home-market. Or course, if Apple announce a ZFS-based upgraded Time Capsule/Home Server at MacWorld ExPo tomorrow that’ll happen sooner.

From a capacity perspective, with 2TB or greater drives being the de-facto standard capacity by the end of 2009 (compared with 1TB today), the growth of all types of media; including photographs, personal video and the increasing availability of internet-distributed hi-def content, coupled with the pack-rat nature of most of us (me included), demand for storage capacity has never been higher. It’ll also push the more mainstream storage user towards the 10-12TB Unrecoverable Read Error issue , aka known as the death of RAID5. 

To deal with increasing capacity and the straight line graph of bit-error rates, drive manufacturers keep making their drives smarter to handle errors and attempt to minimize data loss. This is the wrong approach, but it’s unavoidable as otherwise they commoditize themselves further. Drives should be stupid and let something further up the stack manage this. That something is ZFS.

Of course, this mostly applies to cheaper SATA drives. The more expensive UltraSCSI and SAS commanded a premium for performance and reliability. Move reliability into the filesystem and you’re just paying a premium for performance. Obviously, some need the performance which is why these drives and ancillary equipment and technologies like Fibre Channel will stay around, but I think it’s worth considering if you really need that CLARiiON or Symmetrix…

I have to say, Valorie in Drobo tech support did a great job handling my case. It seems that it may be a PSU problem (yes, just like my Lacie drives!). Both the Drobo and the Lacies come with cheap Chinese-made PSUs.

As you can see from the photo, it should be capable of outputting 6.7A at 12V. My multimeter said 12.36V, though I’ve no way of measuring the current draw when active. Drobo support said that the 1TB Hitachi Ultrastar HUA721010KLA330 drives are a known “power hog”. The drive itself states a current draw of 850mA (I have 2 installed, the other 2 are 750GB Seagate Barracuda ES ST3750640NS, with a stated draw of 520mA, so there could be some truth in that. If this is the case, I know from personal experience that it’s possible for multiple restarts to corrupt volume information resulting in data being lost.

They are sending me a new PSU, so we’ll know soon enough.

I’ve not had time to jot down my reasoning behind this, but in summary this configuration gives me the capacity I need now, room for future expansion, fits into my rack at home, and is (almost) in budget.

I’m a reasonably demanding home storage user, I guess I have around 3 TB of data and 4-5 TB of capacity at home. Photos take up close to 1TB, I’ve a lot of my own video, transcoded video, and music.

I’ve tried all sorts of solutions – external drives, Infrant (now NetGear) ReadyNAS, Drobo, etc. They all have their shortcomings. My Drobo, with over a TB of data let go. My ReadYNAS 600 died of old age (it was sad to see it go). Playing with multiple 1TB external drives is fragile and stressful. All of these products (except the ReadyNAS) are black boxes, the most status you get is a green LED or no mounted volume. At least the ReadyNAS has a MIB available, and I’ve used Nagios to monitor my ReadyNAS NV:

I’d also been thinking about failure, and I’ve had controller failures, PSU failures, disk failures; and it occurred to me that even RAID5 may not cut it. It seems I’m not alone in thinking this, and I found that Robin Harris over at storagemojo.com had been thinking this since July 2007. He also thought some more about disk reliability. I also like Joerg Moellenkamp’s presentation about the end of RAID5.

Feeling validated in my concerns, I decided to brainstorm a little and decided I need to have a storage facility to provide the following:

Fast response and transfer time; simply locate it on my GigE LAN at home.
Around 12TB storage, expandable
Rack-mountable
Open Operating System
Very configurable
Monitorable
The ability to support something more reliable than RAID5.
Be affordable, budget around $2k

There’s only one thing that meets all these criteria, I have to build my own ZFS Server.

By the way, fsck is not a rude word!

Firstly, I had a few hundred GB of ISOs that I needed to off-load from some external drives (for temporary storage, I use and recommend the 1TB Samsung SpinPoint F DT HD103UJ, a nippy and (so far) reliable drive with a 3-platter array. Maximum PC has a good review. I combine it with with good looking and performant Icy Dock MB559UEB-1S-B, with USB2 and Firewire 800 interfaces. I moved the ISOs over to the Drobo, after some testing, and of course, after I’d deleted the ISOs from the external drive, the Drobo started playing up.

It would restart a few times a day, and then it would hang occasionally. After leaving it for a while, I’d always have to cycle the power by pulling the PSU lead out from the back. I’d opened a support ticket with Drobo when the thing stopped mounting.

Something clearly had become corrupt, as when I tried to mount the Drobo volume my Mac (both my Mac Pro and my Macbook Pro), would crash with the black screen. Not good. I used a combination of tools, including Data Rescue II, Techtool Pro, and Disk Warrior 4 – they couldn’t help. At this point I knew Drobo couldn’t help, and as the data was recoverable (albeit in a time-consuming way), I blatted the hardware and drives and put them in a factory default condition. I’ve upgraded the Firmware to v1.3.0, the dashboard to v1.2.4. and guess what, the damn thing is still restarting.

I’ve had it with consumer grade crap. I’m building a ZFS server.

I had some drives from my failing ReadyNAS 600 (now retired), so after temporarily copying the data to a non-resilient drive I loaded the following in the the Drobo and copied the data onto it.

Slot 0: 1TB Hitachi HUA721010KLA330
Slot 1: 1TB Hitachi HUA721010KLA330
Slot 2: Seagate 750GB ST3750640NS
Slot 3: Seagate 750GB ST3750640NS

These are good quality SATA 3.0 Gbit/s drives that I even ran through SpinRite (highly recommended) to check for errors.

Anyway, I’ve discovered that when I unlock my Mac Pro (the screensaver automatically locks after a while, but I don’t put it to sleep), the Drobo restarts. No warning, just a whirring reboot. I’m running the 1.2.2 firmware and the 1.2.1 dashboard. I’ve tried plugging it into another Firewire port and also unplugging my other Firewire devices with no luck.

I’ve logged a support ticket with Drobo, let’s see what they come up with.

I’ve used JungleDisk for ages, very happy with it for backup purposes and mobile, secure access to my files.

So I was more than interested to discover that JungleDisk has released JungleDisk Desktop v2.0. I was particularly keen to see one improvement, the ability to access multiple buckets at the same time.

It’s a great service, cost effective (for me at least – I’ve indefinitely postponed buying a new NAS unit) and reliable.

Highly recommended.