DIY Storage Part 3: Pricing Pitfalls

This is the third installment in a series on building your own SAN. In part 1, I talked about why I decided to build my own array and outlined hardware requirements. In part 2, I discussed how I chose Windows Storage Server 2012 as the software for the SAN.

In this installment, I'm going to talk about how vendor lock-in increases the cost of storage by preventing the use of commodity drives. The fact is, with commodity components you can get a full-featured storage system that can run serious workloads for half the price of a name-brand system.

Take direct attached storage from a tier-one vendor such as Dell. First, an MD1220 DAS unit runs about $4,000 dollars without discounts. I can get the same features in a DAS storage array from Supermicro for about $2,000 dollars. The Dell MD1220 is SAS only, whereas the Supermicro JBOD can run both SAS and SATA disks.

Let's say I want to put some "slow" higher capacity storage into this MD1220. If I purchase a 1TB near-line SAS drive at 7200 RPM from Dell, I can expect to pay $384 without discounts.

I can buy the same drive from Newegg.com for $259 or from my Supermicro dealer for $229. However, I can't use this commodity drive in my MD1220 DAS box because it lacks Dell firmware.

Beyond basic costs, there are other considerations when trying to budget creatively. Near-line SAS drives sold by brand-name vendors aren't going to use anywhere close to my 6Gb/sec SAS2 interconnect speed, so why can't I consider SATA drives? Sure, SAS is dual-port 6Gb/sec while SATA3 is only single, but I can use interposers if I want to retain failover capability to a secondary controller.

More importantly, if the MD1220 supported SATA, I might be able to look at a drive by a different manufacturer and get a much better deal. How much better? In the higher-end commodity SATA drives from Western Digital or Seagate, I might pay only $150 dollars for the equivalent in SATA3.

But what if I wanted a little bit faster drive? Why shouldn't I put a 10K RPM SATA disk in at the same capacity, perhaps a Western Digital VelociRaptor like the WD1000CHTZ? This drive costs about $300 dollars from Newegg and about $260 elsewhere.

Not only is it substantially cheaper, its I/O characteristics are nearly double the near-line SAS disk offered by Dell (200Mb/sec sustained) versus 112Mb/sec on the near-line SAS disk with the same amount of cache (64GB) and a much faster seek time. Western Digital provides a five-year warranty with this drive, which provides confidence in drive integrity.

The interesting thing about using a 10K drive that actually costs less is that if I buy a JBOD that supports SATA as well as SAS, and stock it full of VelociRaptors at 10K in a standard RAID 6 array with a hot spare, the class of the storage I'm buying (for less money) is entirely different than my 7.2K NLSAS drives.

Instead of having super slow archival drives, I'm effectively doubling my spindle count and creating a light- to light-moderate virtual machine data store for less money. I could run lightly utilized Microsoft Exchange data stores, or even a light SQL workload, on a 20-drive, 10K RAID 6 array.

Consider Controllers

But there's more to this story than getting better bang for your buck on SATA drives. Let's talk about the controller that is going to drive this JBOD. If I'm attaching this JBOD to a Dell server of a recent generation, chances are that I'm going to be using an H7XX raid controller to drive it.

Dell's H7XX series raid controllers use either the LSI 2108 or 2208 I/O processor (depending on generation). These are the same I/O processors that are offered in high-end LSI RAID controllers. I'm glad Dell rebrands a quality RAID controller, but there are some serious caveats to consider.

Unfortunately, because the H7XX RAID controllers use Dell firmware, I can't take advantage of two really neat features of the true LSI controllers, namely CacheCade and FastPath. Why do I care about these features?

Well, if Dell allowed me to use SATA drives in the MD1220, I could put in an Intel S3700 series SSD into my JBOD. Instead of paying $1,293.00 dollars list price for a Dell-supported "value" SSD, I could pay around $400 for one with the same capacity.

Why would I want to put a single SSD into my JBOD? Because with LSI's CacheCade (a $160 software licensing option that's built in to the best controllers) I could designate that SSD as a 200GB read-write cache for my rack of slower drives.

In the case of the 10K RPM Western Digital SATA drives, which aren't actually that slow, I could use one or two 100GB S2700 series Intel SSDs as a CacheCade pool to support 22 SATA disks, for less than $200.

This would transform my light- to light-moderate data store into a full-featured, medium-intensity storage pool capable of running serious workloads at half the price.

In my next column, I'll discuss redundancy at the component level and how to distribute storage like this to rest of the infrastructure.