AZ Lamps #18 — Cold Storage Tier

File management is a crucial part of a system’s design. Each piece of data has its lifecycle and is being used differently. For example, photos in a social network frequently appear in feeds and profile pages and can be deleted or edited. On the other hand, a receipt file generated by online banking will likely be downloaded only once.

Azure provides several storage account tiers — Hot, Cool, and Archive. It’s best to compare them by the following parameters:

• Price for 1 Gb of volume.
• Cost per read operation.
• Cost per write operation.

The Hot tier has the highest data volume cost but the cheapest read/write operations. As you might have guessed, it is suitable for “hot” data in active use — e.g., videos or social network photos.

On the other hand, Cool is more democratic in cost per gigabyte (25% off) but has twice the price per read/write. Choose this tier if the data is used occasionally. A good example is tickets or payment files.

Finally, Archive has the lowest cost per volume (20 times less than Hot), but worth keeping in mind a caveat — the files won’t be accessible in real-time. This is because the data goes to magnetic tapes, meaning much physical work is under the curtain. Usually, it takes several days to rehydrate the data from the Archive tier. Conversely, files are stored on fast SSD/HDD disks within the Hot and Cool tiers; thus, they are always online.

The Hot/Cool/Archive trio has been there for a while, but it will soon transform into a quartet — Azure released a preview of a brand new Cold tier. This guy sits between Cool and Archive, closing the gap between the two and extending the choice for developers. The Cold tier aims for such scenarios when the file is written only once and is not touched anymore while it still must be online (for example, backups). I want to emphasize that the Cool tier never fits here due to its relatively high storage price.

I will use the new Cold plan to store backups and logs when it goes to GA. This will be an excellent contribution to cost saving. Also, I expect Cool will remain for more sophisticated scenarios when data needs to be touched at least a few times during the lifecycle.