My first thought this morning having read a few things about Amazon Glacier is that yes, Glacier is indeed an object storage system using tape as it’s long term retention media.
Front ended with disk to provide the landing area for data transmitted via the Glacier object interface, keep inventories, handle replication, and act as the 24 hour download area. When you’re waiting for your 3 to 5 hour time to first byte window to pass that’s a robot picker waiting for a tape drive or drives to free up so it can load the required tape volumes and start reading them off to a disk staging area from where you then download your data.
Nice image, though I expect the pickers will never be idle for any real amount of time.
Now since the words immutability and encryption were mentioned we’ll assume the encryption happens at the smallest level of abstraction, the multi-TB archive, and isn’t done using the tape drives themselves. Those tape drives are also writing to WORM tape to provide the immutability so deletions could be a case of shredding the unique archive object encryption key.
Get rid of the key the data was encrypted with and to all intents and purposes it’s been deleted. Even if the media it was written to is WORM and no matter where any replicated copies inside Glacier are stored.
There are no real secrets in IT, there are just things you haven’t heard about yet and what I think we will hear about soon enough is that Glacier is ‘cloud tape’. An object namespace with tape as it’s long term storage media.
Could it be disk end to end? Yes. It’s viable if you’re willing to use the barest of metal and Amazon have never had a problem living on margins which are tiny bits of nothing. Maybe it’s aggressive compression and decryption which requires hours of CPU cycles before you are presented with something you can then start downloading across the internet.
We’ll know more soon enough.