Iain Rauch's Blog

We already know we’re using one of those eBuyer value cases as a starting point. The first thing to do was strip out all the internals – we just needed the bare case. It was pretty simple to dismantle as almost all of it unscrewed.

Next, the reinforcement I mentioned in the previous post. This was a 3mm x 25mm steel strip which was cut and epoxied to the bottom. Strategically placed, they raise the PMPs off the bottom slightly. Combined with 20mm standsoffs there is enough space for the connectors underneath. Those are just standard PC-modder parts wired together as specified in the diagram from the last post.

The drives are not screwed in, they just rest there – connectors down. To prevent them from falling over, there is a grid at the top. This was made by combining 1mm x 3mm steel strips with some dowels. Strangely, the dowel of this size was difficult to find at a reasonable price, so we ended up using 2mm diameter polyester fibreglass rods. For vibration dampening, the only thing is rubber washers securing the PMPs.

Holes in the front for the fans and the case is complete (almost). Unnecessary, but for the cool factor we wired up LEDs for each drive. The PMPs come with pin outs for the PMP status as well as each drive. Quite a lot of effort to connect so many LEDs (the right way round), but a good indicator when switching on the case.

First and foremost, I want to thank my friend Mat for helping with the design. I couldn’t have done it without you!

As you will recall from part IV, I had a second 4U case used only to house drives. This was the starting point for my NAS server 2, case #2. I would have one case for the host system (and some drives) and a second case just for drives.

If you look at the Backblaze design, you’ll see it has space for a motherboard, 2 PSUs and 46 drives (1 is for the host OS). That’s an awful lot to squeeze into a single case! It works for them as they are racked up in a data centre with deep racks. Something would have to give in my 545mm deep ‘value’ case.

Fourty-five drives is a nice number, as it gives you 3 x 15 drive arrays. I think that’s the most it’s sensible to have in a RAID 6 configuration. Fitting up-to 11 drives in the first case, I needed to cram at least 34 in case #2.

Having verified SATA port multipliers work perfectly well, I was happy to take the same approach as Backblaze with their backplanes. What I decided to think about a bit for carefully than them was the host SATA ports. They put 20 drives on the PCI bus, but I knew from experience this would degrade performance.

Since I already had a 4-port PCIe card, it made sense to use that for 4 port-multipliers. This would give me 20 ports on a PCIe x4 bus – the drives won’t all operate at maximum speed at the same time, but there’s still plenty enough throughput to saturate a gigabit LAN. This puts the SATA port count at 26 so far. Looking for 45 in total, I would need another 19. The only practical way to do this, based on the number of PCIe slots I had was another 4-port card each hosting a 5-port multiplier/backplane. So instead of cramming one standalone port-multiplier and associated drives in the main case, we decided to put them all in case #2.

Spanning power between cases didn’t seem like the best idea, and with 40 drives now needing to be catered for, 2 PSUs had to fit in the second case. Here’s a bird’s-eye and front on view of the layout. Thanks again, Mat, for coming up with it.

The black bars going left-to-right are for reinforcement – 40 hard drives weigh a lot! The other thing to note is the orientation of the drives. Keeping them that way allows efficient air-flow, from front to back.

The original intention for my 1000W PSU was to power the whole system and 22 drives. Since I’m now trying to support up to 45 drives and keep the power in each case separate, the power arrangements needed to be rethought.

I allocated the OCZ PSU to case #2 and re-purposed an old PSU I had for the main system. I calculated the amperage available for each voltage sufficient to supply 5 of the 8 port-multipliers. Maybe it would have been ‘neater’ to split it four and four, but this way, when upgrading, I wouldn’t need such a beefy supply.

Each backplane requires two molex feeds. Here’s a wiring diagram (credit once again to Mat).

Now we know pretty much what we’re building, in the next post I’ll talk about the parts and construction.

Most likely I found it on Engadget, but it did crop-up in quite a few other places. I am of course talking about the Backblaze blog post ‘Petabytes on a budget: How to build cheap cloud storage‘.

When I found this, I thought it was fantastic – 45 drives and a host in a single case. If anything was ever going to be perfect for me, this was it!

In the first post, Backblaze were kind enough to detail all the parts used and make available a 3D model of the enclosure. This was great for the community, but in a follow-up post they directed readers as to where they could order the enclosure from directly – Protocase.

As their website directed, I emailed them straight off to get a quote. $872 – wow, way more than I was expecting. On top of this, it would have to be shipped from Canada, adding to the price substantially. Dismayed with that outcome, I thought it was the end of the matter, but Protocase emailed me a few days later asking for feedback.

As you do (or at least as I do), I replied with a 250 word rant as to how expensive it was compared to products such as the Norco RPC-4020, which was a mete $280. It must have be somewhat interesting, as I got an even longer reply direct from the Chairman addressing each of my points.

Needless to say, I didn’t go forward and purchase one of these cases, instead using the concept to design my own.

In my next post I will talk about the design along with how and why it differs from the Backblaze storage pod.

Wow, sorry for neglecting this series for so long – I got distracted by being employed! Hopefully, it’s not so long that I’ve forgotten my trail of thought.

So at the end of my last post, I had four 1.5TB hard drives in RAID 6. That would have been around December 2008, but come January 2009, I was out of space and needed to add some more. This was easy; I had 11 bays in the case, and six on-board SATA ports – I just added two drives and connected them straight up. This gave me another 3TB of useable space without any hassle.

In June 2009 I needed to upgrade again, but this time, things were a little trickier. I had no on-board ports left so had to decide how to expand. The original intention was to use 8-port PCIe cards; with space for two of these, I’d have ended up with a 22 drive maximum.

Now, I’m not sure exactly what my thought process was at the time (don’t forget this was two years ago) but I probably decided 8-port cards were either too expensive, or just wouldn’t get me enough ports in total. I ended up getting a 4-port card and four hard-drives to go with it. Great, another 6TB in the array and I was happy until October.

So what did I do next? I’d used 10 out of my 11 bays and had no more SATA ports left. [Probably] being desperate for space,  I just ordered another of the 4U cases that hold 11 drives. Seeing as how my 4-port PCIe card supported them, the cheapest way to get extra SATA ports was to use a SATA port-multiplier. I gave it a go and £40 got me 5 ports, but obviously I had to sacrifice one from the PCIe card.

The PMP was very successful, although I did have to disable NCQ to get it stable. This isn’t necessary anymore, so I won’t go into any further detail. Just to keep track, at the end of October 2009 I had 12 x 1.5TB drives in my RAID 6 array.

In the next installment, I’ll explain where the insparation came from.