[latchkey]

I don't have a personal need for this right now, but the fact that it stores the data as EBS volumes is pretty cool. I could imagine having local servers automatically mirrored so that they could be failed over to instances on ec2. Very powerful stuff indeed.

[cperciva]

Note that the mirroring is asynchronous -- if your local server fails you can replace it with an EC2 instance, but you lose anything since the last snapshot.

[kondro]

Not since the last snapshot. It syncs as close to real-time as possible without the latency and network speeds affecting the performance locally.

This solution isn't designed to replace fault-tolerance on local hardware. It is for close to realtime offsite replication and backup.

Data in S3 is stored in at least three geographically separate locations and snapshots are very fast and very efficient on storage space.

The final major advantage you get through a solution such as this is that if you do have your primary site go down (floods, tornados, etc), you can bring up all your existing images via EC2 without having to have a bunch of redundant hardware sitting around waiting for disaster to strike.

And what do you pay for this? $125/month plus a per GB storage cost CHEAPER than enterprise storage generally is.

[sciurus]

Let's say I need 50TB of usable space. I can purchase an Equallogic PS6100E (a mid-level "enterprise" storage device) with twenty-four 3TB drives and 5 years of support for $85,000. Rack space and power isn't free, so let's say the total cost of equipment and facilities over 5 years is $100,000.

In contrast, storing 50TB for 5 years on S3's reduced redundancy storage would cost $250,000. If I ever need to transfer any of that data back to my data center, there'll be a hefty bill for that as well.

[kondro]

Except to get the same redundancy as this AWS product you would need a replica of this storage device in at least 2 other datacentres, you would have to ensure you have enough spare capacity lying around for any snapshots you take and have to buy or build software that manages this 3-way data synchronisation between sites.

Also, you seem to be forgetting that you still need local storage for your images anyway. This is a hands-off backup and disaster recovery product.

For full disclosure, the Storage Gateway's pricing isn't the same as S3's at the moment. They only have one storage tier and it is $0.14 per GB, no discounts. Therefore, 50TB of storage over 5 years is $420k.

Having said that, what would it cost to:

* Not only have your primary but 2 x secondary PS61000E's.

* 2 extra datacentres with connectivity to themselvse and your primary site.

* Software to manage asynchronous streaming of data from your primary to 2 x secondaries.

* Software to take consistent backups of these images and store in at all three locations.

* Software to ensure that your secondary sites contain only encrypted data.

* Cold-spare hardware at a secondary site capable of running all your images.

[evandena]

With Amazon you could facilitate a DR onto their other platforms. With your Equallogic example you would need to mirror that data to another device with recovery targets in another location.

[nknight]

At the scales he's talking about, 10gbps uplinks are easily available. That said, the killer isn't the equipment costs, it's the people.

[ridruejo]

If I read it correctly, the server can fail but because the data is stored via iSCSI, it will not actually be lost (unless it is the appliance itself that fails)

[cperciva]

Sure. But if we're ignoring the case where the appliance doesn't fail it's no better than a drobo.

[tiernano]

the storage for the appliance can be outside the appliance (DAS or SAN storage). the appliance can die, but you just plug in a new appliance and point at the storage, and your back... as for the drobo comment, this should be faster than a drobo... and even ignoring the appliance fail, the off site backup is key.