Norco DS-1210 + Mac mini

I recently dumped my old PC home file server in favor of a Mac mini. I had been wanting to switch to a Mac based server for some time, mostly easy administration and standardization on the Mac as our home platform, but the problem of disk space and multiple drives for RAID on the mini is obviously an issue and the only choice after that is a Pro which is completely overkill for my needs. The old PC could fit four drives easily and I wanted to handle more than that so I was looking at external FireWire cases, most FireWire external enclosures handle 1, 2, maybe 4 disks in an annoying oddball form factor with external power supplies which translates into a mess when you have a few of them. The multi-drive cases also seemed completely overpriced for what they were. I had been looking at stacking 1U FireWire enclosures but they seemed excessively expensive for only 4 disk slots, more if you want hot swap. I don't mind spending money, but I do mind not getting value.

I finally ran across the NORCO DS-1210 and was instantly sold, 12 PATA drives in a rack-mount FireWire enclosure with hot-swap trays, $569.11 delivered, perfect.



I picked the unit up from www.excaliberpc.com, bought the Mini and luckily enough Outpost.com was having a sale on 300GB drives so I picked up four to add to the two I already had, not to mention the pile of smaller drives I could throw in. Total cost for the new terabyte server, 569.11 (DS-1210) + 586.18 (1.66 mini) + 299.96 ( 4x300GB) = $1455.25.





The DS-1210 arrived quickly and is quite the box of steel, I powered it up and my daughter asked if it was an air conditioner. Definitely designed for a noisy server room, it'll be going in the basement so that's not a big deal and I might put some quieter fans in it. The only downside on arrival was two screws floating around inside which were supposed to be holding part of the front LED status panel in place, ah well, nothing a screwdriver can't fix. Also, the IDE cables weren't necessarily in the right orientation, half of them had to be flipped in order to get a drive in the tray, again, no biggie. The unit comes with a FireWire cable, PCI FireWire card, drivers, and screws to mount the 12 drives, so it is a complete package.







Some might balk at the excessiveness of hot-swap in a home environment, after all I have only had one drive fail me in 10 years and within warranty. That's not really the point. Hot-swap makes it easier to do things like backup to a hard disk which is taken out, far easier to upgrade disks on the fly, easy to perform recoveries from dead machines:


and so on. One of the first things I did was go through a small pile of old drives and do a multi-pass erase on them. Far easier to deal with the trays than messing around with cases. The icing on the cake was OS X's ability to read NTFS disks so I could easily take anything I wanted to off my old drives before wiping them.

As mentioned the unit has 12 drives, but the drives are on 6 IDE channels, so the drives are configured as master/slave/master/slave ... across the front. This means a bunch of things:

A - The instructions say configure the drives as master/slave, of course I had to try cable select and it does not work, so all your drives must be master/slave if you want them to work. Easy enough.

B- When you pull one drive, the IDE channel goes down, so really two drives go down.

C- Configuring software RAID sets with drives on separate channels is slightly faster than both drives on the same channel.

E- Configuring software RAID sets with drives on separate channels means that when you pull a drive in one set it will take out a drive in another set (see B).

This all means you have to plan your RAID sets a little more carefully than if the drives were truly individual hot-swap, especially if you plan on pulling drives while volumes are mounted.

The unit itself has a switch for RAID 0 (striping) which treats all 12 trays as one drive. After setting up my drives and putting a lot of stuff on them I started to have nasty thoughts that the switch would get accidentally turned on and one or more of my disks would get screwed up. I took the RAID 0 switch out entirely and put a jumper in, I feel better now.

I decided to configure my sets as mirrored pairs on the same channel, while this is the slowest and will take the whole set offline when a disk is pulled, it is the simplest to administer. If you have multiple RAID sets on the same channel when you pull a disk you affect more than one RAID set, and if these sets are mounted you end up effectively downgrading more than one RAID set and having to rebuild more than one disk. The other small thing I set up was a 4MB Unix partition on each disk named the slot number the disk was in, this would make it super easy to identify which disk was where.

Since this is a home file server it does not need to be brutally fast, it is mainly for backups and storing audio, video and family pictures, nothing too demanding speed-wise but it does demand a lot of space. Anyway, let's take a look at the performance.

I used the disk benchmarks in XBench with a pair of 300GB Seagate Barracuda's, I did four runs for each configuration and am using the best numbers out of all the runs. Overall the runs were extremely consistent with an inevitable slow number sprinkled about probably due to some other system activity.

The results are pretty much what you would expect for the configurations and the use of software RAID.

All numbers are MB/sec

DRIVE 0, Master, Slot 5

Sequential Uncached Write 4k blocks:  25.34
Sequential Uncached Write 256k blocks:   22.43
Sequential Uncached Read 4k blocks:   09.98
Sequential Uncached Read 256k blocks:   27.67
Random Uncached Write 4k blocks:   00.97
Random Uncached Write 256k blocks:  18.66
Random Uncached Read 4k blocks:   00.67
Random Uncached Read 256k blocks:  17.47

DRIVE 1, Slave, Slot 6

Sequential Uncached Write 4k blocks:   25.34
Sequential Uncached Write 256k blocks:   22.43
Sequential Uncached Read 4k blocks:   09.88
Sequential Uncached Read 256k blocks:  27.63
Random Uncached Write 4k blocks:  00.85
Random Uncached Write 256k blocks:  18.57
Random Uncached Read 4k blocks:   00.68
Random Uncached Read 256k blocks:  17.41

This shows both drives performing almost identically and regardless of slave or master.

Drive 0/1, Master/Slave Mirror, Slots 5/6

Sequential Uncached Write 4k blocks:   13.97
Sequential Uncached Write 256k blocks:  11.35
Sequential Uncached Read 4k blocks:  09.43
Sequential Uncached Read 256k blocks:  23.54
Random Uncached Write 4k blocks:  00.64
Random Uncached Write 256k blocks:  09.86
Random Uncached Read 4k blocks:   00.68
Random Uncached Read 256k blocks:  12.75

Drive 0/1, Master/Slave Mirror, Slots 5/8

Sequential Uncached Write 4k blocks:  14.85
Sequential Uncached Write 256k blocks:  12.11
Sequential Uncached Read 4k blocks:  09.21
Sequential Uncached Read 256k blocks:  27.68
Random Uncached Write 4k blocks:  00.88
Random Uncached Write 256k blocks:  12.00
Random Uncached Read 4k blocks:   00.68
Random Uncached Read 256k blocks:  19.28

This shows that using separate channels for a RAID 1 set is faster than using the same channel. This also shows that a software mirrored pair is considerably slower at writing than an individual drive.

Drives 0/1, Master/Slave Stripe, Slots 5/6

Sequential Uncached Write 4k blocks:   27.80
Sequential Uncached Write 256k blocks:   21.89
Sequential Uncached Read 4k blocks:   9.32
Sequential Uncached Read 256k blocks:  27.04
Random Uncached Write 4k blocks:  1.27
Random Uncached Write 256k blocks:  17.19
Random Uncached Read 4k blocks:   0.71
Random Uncached Read 256k blocks:  12.91

Drives 0/1, Master/Slave Stripe, Slots 5/8

Sequential Uncached Write 4k blocks:  29.57
Sequential Uncached Write 256k blocks:  23.65
Sequential Uncached Read 4k blocks:  09.33
Sequential Uncached Read 256k blocks:  29.38
Random Uncached Write 4k blocks:  01.72
Random Uncached Write 256k blocks:  22.77
Random Uncached Read 4k blocks:   00.71
Random Uncached Read 256k blocks:  19.47

This also shows that using separate channels for a RAID set is faster than a single channel and it also shows that striping is a little faster than a single drive in most cases.

In general it appears that the unit as a whole has good FireWire performance and the drives and software RAID are more of a limiting factor.

If you have somewhere to put a noisey machine and want a lot of storage on the cheap with decent performance, this is a pretty compelling setup.

The only thing left to think about now is how to put the mini inside the DS-1210...