(or "What the heck is ephemeral storage?")
The EC2 "Hardware"
When an EC2 instance is launched, a virtual machine is reserved for the instance to run. That virtual machine has particular specifications depending on the instance type: 32-bit or 64-bit CPU, number of virtual cores, size of hard drive, etc. Details about the instance specs is available at http://aws.amazon.com/ec2/#instance.
When your EC2 instance is in a "running" state, that means that it’s running on the virtual machine, and this is what you get charged for.
The virtual machine’s hard drive is considered "ephemeral". The term "ephemeral" comes from the greek word "ephemeros" which means "lasting only one day". Anything on such a hard drive should be considered temporary. Unless the data is copied off the hard drive, if the virtual machine is stopped, then the data is lost. This includes data, software, and even an operating system that resides on those hard drives.
Amazon Web Services provides EC2 instances with two types of root devices.: "EBS-backed" and "instance store".
"Instance Store" Instances
An "instance store" instance is an EC2 instance whose root device resides on the virtual machine’s hard drive. When the instance is created, the base AMI is copied to the virtual machine’s hard drive and launched. The instance can run for as long as you want, but it cannot be stopped. Since the instance’s root device is the actual hard drive, it is "stuck" on the hardware, and the only thing you can do is terminate the instance. If you do this, the instance is deleted, never to be recovered. You also run the risk that if the virtual machine’s hardware fails, then you will also lose anything on the hard drive.
If you launch an "instance store" instance, be prepared to leave it running until you’re completely done with it. Note that you will be charged from the moment the instance is started, until the time it is terminated.
An "EBS-backed" instance is an EC2 instance which uses an EBS volume as it’s root device. EBS volumes are redundant, "virtual" drives, which are not tied to any particular hardware, however they are restricted to a particular EC2 availability zone. This means that an EBS volume can move from one piece of hardware to another within the same availability zone. You can think of EBS volumes as a kind of Network Attached Storage.
If the virtual machine’s hardware fails, the EBS volume can simply be moved to another virtual machine and re-launched. In theory, you won’t lose any data.
Another benefit, is that EBS volumes can easily be backed up and duplicated. So you can take easy backup snapshots of your volumes, create new volumes and launch new EC2 instances based on those duplicate volumes.
Probably the biggest advantage "EBS-backed" instances have over "instance store" instances is that they can be stopped. When you do this, the virtual machine is shutdown and the EBS volume is stored for later retrieval. The hardware is then available for someone else to use. In addition, during this time, you are not charged the EC2 instance running charge. But you are charged for the EBS storage. When you want the instance to run again, you just start it up again. A new virtual machine is reserved, your EBS volume is attached, and your instance is booted.
But what about the virtual machine’s hard drives?
Yes, it is possible to use those hard drives, even when your EC2 instance is "EBS-backed". By default, they are not available. If you use the command line programs to launch your instance, you can use the "-b" option on the ec2-run-instances command to attach the "instance store" drives to your EC2 instance.
Having these drives available can be beneficial if you want to store temporary data. Read and write access should be faster than reading from and writing to an EBS volume because you’re not sending data over the network. In addition, you won’t be charged for data transfer or data storage. But this only works if the data can be lost at any time.