I spent yesterday in a training class for the Isilon system that we use at work. Isilon offers a clustered storage solution that is quite impressive both in its performance and its implementation. I know that you can go the web site and look at all the specs, but I thought I would take some time and share the things that jumped out at me during the first day of training.
First off, I experienced a slight culture shock when they started talking about capacities. Speaking from a developer's perspective, I'm used to talking Megabytes when referring to my applications and possibly Gigabytes when talking about data storage. These guys talk Terabytes and Petabytes. Imagine what you could do with a petabyte of code. I think Skynet became self aware right around 1.3 petabytes.
Now as you can imagine its pretty hard to manage that much data. The mainstream file systems just won't cut it, so they wrote their own. They based their OneFS file system and OS on FreeBSD. Each node in the cluster runs the OS and communicates with the other nodes so that each node knows what the other nodes have and are doing. This way, there isn't a single controller unit. Any node could go down completely and the system will continue to run without a hiccup. It also makes it easier to add expansion nodes. You can easily add several terabytes of new storage to the system in approximately 60 seconds.
One of the coolest decisions that they made was to stripe files across nodes and not across disks(most nodes have 12 disks). This provides the highest level of data protection in the event of a failure.
In today's class, we'll be looking at some troubleshooting exercises and hands on labs. Should be fun.