In a previous post, I discussed building a very large storage array for under $5000. Since the time I have written the original article, prices have continued to come down, drive capacities went up, iSCSI target software became more stable, and many other things that make this even more attractive than ever. My plans are mostly unchanged from the previous post, but I thought it would be relevent to summarize them here so you don’t have to read that whole post.

I think it is noteworthy to say that there are other drives out there, some cheaper and some larger. I wanted to go with the ES.2 (or WD RE) as they are designed more for 24×7 operation and this type of application. Using consumer level drives for a setup like this is generally a bad idea. It should also be noted, that if you use Western Digital drives, you should use the RE version as opposed to the SE version. Trust me on this. The RE version is RAID enhanced, which has several differences between the SE, but the most important being that the RE is firmware encoded to be the given size. An SE 500GB drive may be 499.7, 501.2, etc. The RE 500GB drive will always be 500GB. This is very important in a RAID setup, where replacing a faulty 501.1GB drive with a 499.9GB drive will NOT work, be very frustrating, and probably make you want to hang yourself.

Additionally, there is other, more chocolaty controller cards. If you have something against 3Ware, I recommend looking at the Areca cards (they even have a nifty LCD panel to show the health of the array, perform rebuilds, etc.) I chose the 3Ware from personal preference. I have used hundreds of their cards and never had problems. Why fix what isn’t broke.

This setup, based on which RAID setup I choose, will yield the following storage space:

The technical nitty-gritty is beyond the scope of this post, but here is a summary of why I chose RAID50.
Because the majority of my IO operations to the array are read operations, I wanted something that had a bit higher performance on the read operation, which I was willing to sacrifice a bit on the write operations. RAID 0 provides the most read operation performance boost, however it entirely lacks any fault tolerance. As you increase the number of drives in your array, the chances of a single drive failure increase. In a RAID0 setup, any 1 drive would destroy the entire array.
So… Because Fault Tolerance is a requirement (I’m not going to re-encode 5TB of video unless I have a catastrophic fire and cant grab the array in time.), that rules out RAID0. RAID 10 is 2 mirrored RAID0 stripes. This is a rediculous waste of storage capacity without any major benefits, other than you get the performance of RAID0 and the Fault Tolerance of RAID1. Scratch RAID10.
RAID 6 implementations are rare and not well documented, but it is basically RAID5 with dual parity. I scratched this because its confusing and offers no major advantage over RAID50.
RAID 5 offers enough fault tolerance for my needs, with a slight performance it.
RAID50 offers the same great redundancy as RAID5, with a read performance boost like RAID0. It takes a slight penalty on the write operations because it must write parity information more than once. Most of my ops are read, so this one is FTW.

In regards to my server… Yes, I know this is cheap. No, I cant replicate this price exactly and consistently, but eBay has a lot of bargain barrel servers for the cheap. It needs to serve one purpose only, and that is to hold the 16xPCIe RAID controller card and serve as an iSCSI target. Nothing more. I intend to use Windows Server 2008 or Linux using IET. This iSCSI target will connect to multiple other computers and PVR systems with iSCSI initiators via dedicated Gigabit network (more on this later).