Amazon’s AWS group announced a new low-cost cloud-based storage service that offers replicated storage for as low as $0.01 per GB (note: I say as low as because it costs 10% to 20% more in the western United States and outside the United States – but that’s still only 1.1 to 1.2 cents per month per gigabyte.) There are a plethora of other charges as well – for more information see the Amazon Glacier Pricing link.

Here’s what we know about Amazon Glacier so far:

  • It’s not tape.
  • It’s intended to replace tape.
  • Data stored in Glacier takes 3-5 hours to access.
  • It is built upon cheap commodity hardware
  • It offers replication within a single facility.
  • It offers multi-region data replication.
  • It supports 256-bit AES encryption.
  • It has a claimed 99.999999999% annual data durability guarantee per object.
  • AWS import/export is supported for bulk data migration.
  • Multi-part uploading is supported.
  • AWS will eventually offer an automatic migration system between its S3 storage with lifecycle policies.

Amazon’s AWS S3 pricing is uneffected, so it $0.13/GB to $0.06/GB along with I/O costs (Microsoft’s Azure pricing starts at $0.125.)

So how does this impact backup and archiving? I don’t think that there’s any doubt that this makes public cloud backup a more attractive tertiary destination for backup data. In essence, for small and medium businesses as well as some large enterprises, this makes D2D2C (Disk-to-Disk-to-Cloud) more compelling than D2D2T (Disk-to-Disk-to-Tape.) I can easily see managing your last copy of data within S3 and then using Glacier for all retained data that isn’t on-premise.

At the same time, DRaaS (Disaster Recovery as a Service) is growing in importance – so the management of two-levels of cloud storage will become critical (faster cloud-based spin-up storage versus slower archival storage for system-level data.)

With Glacier costing $10 per terabyte per month ($120 per terabyte per year or $360 for three years), the natural comparison is going to be with LTO-5. A 1.5TB LTO-5 tape can be bought for $50, so that’s about $33/TB, so if you don’t assume reuse, then after 3.3 months, the tape drive will break even in terms of cost. Of course, you have to buy the LTO-5 tape drive, which can put you back $1500 or more – so you’re going to have to store a lot of data in order to get a quick return on investment. Also your operating expenditure is going to dramatically be in favor of cloud-based Glacier.

With that said, I see buyers and customers every day who struggle with 1.5Mbps T1 lines – so in small and medium enterprises I don’t see the bandwidth for cheap storage swap occurring all that quickly.

So what do you think – is Glacier is going to kill tape?


  1. Price-wise, Glacier is a no-brainer. The one snag is that there is no simple way to upload archives to a Glacier “vault” as yet, so it’s not exactly consumer-ready, but I suspect that to be remedied quickly as they provide the necessary API calls for Java, .NET and REST via the AWS SDK.

  2. I don’t think cloud backup is going to happen anytime soon, if bandwidth costs don’t drop it will still be cost prohibiting for most customers to adopt.

    It would be interesting to see ISPs offering their own backup/DR services with dedicated backup/replication bandwidth, aside from standard internet bandwidth.

    1. Ricardo: Pure cloud-based backup (what we call D2C, or Disk to Cloud) is problematic for all but the smallest data sets due to bandwidth – as you call out. But think of the cloud as a tertiary backup target – in other words, D2D2C (Disk to Disk to Cloud) – and things like Glacier get more interesting. You now have latency/pricing options in the cloud – so you can have things like D2D2C2C with the two cloud options differing on latency and price.

      Thanks tremendously for the note!

Comments are closed.