Free ESG report : Seamless data management with Avere FXT

Blocks and Files The Register's storage desk has heard that the demand for cloud archival services has grown to the point where Amazon's rivals are considering creating Glacier-like services.

Seagate's cloud-backup subsidiary EVault has been linked to just such an effort, although when El Reg asked, it would not confirm this.

Glacier is Amazon's cloud archive, with cheap per-month data storage rates – $0.01 per gigabyte per month – and slow restores. It is thought to be based on SpectraLogic T-finity tape libraries.

EVault, according to our storage gossips, is going to use disks, next-generation slow and energy-efficient drives from parent Seagate, probably shingled magnetic recording drives, and thus be able to generate restores which are potentially faster than those achieved on Glacier. The service would be compatible with Amazon's S3 interface, say insiders.

But with potentially tens of thousands of drives needing to be kept spinning, power usage will be high. That has to be cut. How?

The Facebook cold storage project uses racks of drives in which the majority are spun down. These are shingled magnetic recording 4TB SATA drives.

Hybrid drives, however – ones with a flash cache as well as disk platters – could be better.

The data would be stored on disk with metadata on the flash. Then disks could be spun down but the flash would still be powered, meaning the metadata could be read.

When writing fresh data to an online archive built from racks of disk drives, you could have just one drive per rack active and have data written to it. When they fill up, you could spin them down and write data to a fresh set of disks, again just one per rack.

Metadata is written to say which files are on which disks. Then, when a restore is needed, the metadata says where desired files (or objects) are located, and the system can spin those disks up. It's a new form of MAID (Massive Array of Idle Drives) technology, and having metadata always available in the flash part of spun-down hybrid drives will make it more workable.

One further wrinkle: the smart hybrid drives could also rotate at slow speed, say less than 5,400rpm, as that could cut power needs and enable a higher areal density, meaning more data stored for fewer watts.

We asked EVault about the topic of a disk-based Glacier-like service and were told, "EVault has no comment on this."

The thing is that tape cartridges in a library take seconds to be located by the robot and carried to a free drive. The effective data-retrieval delivery rate for tape libraries is going to be a lot less than for the same data centre floor space used by racks of disk drives. But tape cost/GB is a lot less than disk cost/GB.

But again, if the total cost of ownership of disk could be brought nearer to tape, meaning acquisition cost plus lifetime power/cooling costs, while delivering more data and delivering it faster than tape, then a disk-based online cold data storage archive might be viable.

If Seagate is indeed developing a hybrid flash/disk drive for cold data storage, it should (theoretically) surely be talking to lots of cloud storage and archival storage product providers. SGI should be in that list, ditto SpectraLogic, as well as EVault – for Seagate will want to ship millions of these slow shingled spinners.

Pretty cold storage MAID arrays, all in a row, might just speed past Amazon Glacier and provide a means to beat the Bezos beast.

Watch this space over the next few months. ®

5 ways to reduce advertising network latency