Do-It-Yourself Fibre Channel Array
skarphedin writes "There's an interesting story here on a do-it-yourself fibre channel array. These guys make one for under $250 and it can perform up there with 15k SCSI in some cases." You know you want one.
1) Buy a lot of fibre
2) Fly to Folkestone, England
3) Rent a sailboat
4) Sail to Calais, France, laying an array of fibre behind you
5) Congratulation on your Do-It-Yourself Fibre Channel Array.
I have used FC tech for several years. It's amazing technology, but performance is not it's main advantage. It's advantage is the possibility of stacking up incredible amounts of storage, with rendundant paths, at up to 100 m from the attachment point (one of the servers). This kind of environment is also very mindful of quality, and a self-made solution is not acceptable. Would you stack dozens of these self-made boxes and bet your career that they'll not fail. I know I wouldn't.
On the other hand, if I just want performance, I will do better with SCSI, and even save some money.
In this respect, I don't quite see what kind of niche would the solution in the article cover.
Sigged!
We had great plans for building an FC array up until a while ago. For those who think FC is too expensive, take a look at this:
180GB ATA drive: $200
Qlogic FC host adapter: $40
10 18GB 10k drives (eBay): $99
10 T-cards: $50
UTP-cable: $20
--------------
Total: $209
Of course, there's the cost of running the array as well, which is the reason we never finished our project (We did get the hostadapter and built a couple of T-cards though). We calculated that our FC array would cost us an additional $2-300 in electricity every year. After getting hit with a $500 surprise electricity bill for our current equipment, we simply decided it wasn't worth it and got another IDE drive instead. Still, an interresting project. =)
Why would one want to use it in a home setup?
For the fun of doing it, for one thing. For instance, I have Token Ring, ATM and serial equipment in my home LAN. Why, when there is FE or even GigE? Because it's fun to play around with, and I learn a lot as well.
Second, you might get pretty darn good performance out of a relatively cheap setup. Modern ATA-drives are pretty fast, but the problem is, there's just a single spindle. Random access will kill your drive. A home-built FC array for the price of an ATA-drive will get you perhaps five to ten separate drives. Mostly, these drives will be 10k drives as well (Almost all ATA-drives are 7.2k or less). The slightly higher rotation speed combined with the fact that you've got a large amount of individual spindles gives you much better random access. Also, remember that ATA usually is a huge CPU hog, which adds to the performance bottleneck.
Where I used to work, we had a few Sun servers with FC disk arrays.
Here's the Sun engineer's explanation of why FC is so interesting for servers:
1) The FC protocol has a 100MByte/s dedicated bandwidth to data. The communication between disks etc. will not interfer with this bandwidth.
2) Modern SCSI has two modes: one for data (burst mode) and one communication mode. The communication mode is a lot slower (first scsi standard) in order to remain compatible with older disks. This means that scsi is a lot more advantageous to users reading large files than small files.
This is where FC becomes interesting: If you have a striped disk array, you will read many small segments from different disks instead of large segments from single disks. In this special case, FC is faster than SCSI, even though it is "slower" by looking at the burst rates in the specs.