Amazon finally delivers on the data-loss on power-down problem

It was always the Achilles heal of Amazon's cloud computing infrastructure, and that was the simple fact, that upon power down, all data stored on the partitions where lost for good.

This meant some creative solutions with mirroring to other machines or periodic updates to their Amazon S3.

Elastic Block Storage (EBS) was then introduced as a way to persist partitions and have them mountable on a running instance. You could take the EBS volume and move it to another machine, or clone it, and then mount the new clone. Point was, it was persistant, even if the volume wasn't mounted to a running machine. You paid for the amount of storage you allocated (not what you used) and the amount of i/o transactions you performed on that.

Amazon have gone the extra mile now with EBS and closed the last hole in their, otherwise, excellent architecture. You can now boot your instances straight from EBS storage. This permits you to mount your root partition (/) on EBS that is then persisted on power off.

Due to the fact the EBS volume is controlling everything, Amazon are claiming faster instance start up times, as there is no local file disk copying having to be done, it will done all over the (one can assume a highly optimized) SAN.

It does allow for some interesting configurations. For example, you can now, shutdown your instance, with all the data perserved, startup a higher specified instance (more memory/cpu) and resume the processing from where you left off. All within minutes. Imagine upgrading your hardware in minutes with no data loss. Very exciting. So that data crunching app can now scale itself up in CPU for the few hours of the day and scale back down again when the daily reports have been run.

This is great news and greatly simplifies the management of instances in the Amazon network. However, it does increase the cost of running an instance, as you still have to pay for the storage and i/o operations on your EBS partition.

Amazon have also announced a new west coast data center where you can choose to have your instances running.

For more information:


Recent Cloud posts

Recent JAVA posts

Latest CFML posts

Site Links