Every time I see this pop up on HN, I upvote. I've been using rclone for years for my main backup [1] (now over 18 TB) from my home NAS to Amazon S3 Glacier-backed Deep Archive, and it is by far the easiest backup setup I've ever had (and cheap—like $16/month).
Egress costs are significant by the time you want to use your backup, around $1800 to download the 18TB, which is around 10 years of storage cost. If you assume 5 years of HDD data lifetime, you need to roughly triple that monthly cost, which is not too bad, but you can buy an enterprise 20TB drive per year for that cost. Personally I would only use AWS in a way that I would almost never have to pay for egress for the entire dataset except when my house gets burned, as a last-resort backup instead of a main one.
I have a primary and backup NAS at home, and a separate archive copy updated weekly offsite locally. Then Glacier is my "there was a nuclear bomb and my entire city is completely gone" solution.
And in that case, assuming I had survived, I would be willing to pay most anything to get my data back.
Nice, that's a great point and a good use case. I normally try to stay away from AWS (when I have to care about the cost) but I think you've found a case where it makes sense!
I've already got backblaze (behind rclone) set up for my backups so adding a glacier archive would be quite easy. Thanks!
Last time I priced it out, Backblaze was more expensive per GB — note that I'm using Glacier Deep Archive, which is something like an order of magnitude less expensive than plain Glacier-backed buckets.
It also incurs a delay of at least 6-12 hours before the first byte of retrieval occurs, when you need to restore. So there are tradeoffs for the price.
Oh, I see. I'm shamefully ignorant about everything concerning backups, so I just googled "Amazon S3 Glacier" and compared some numbers that pop up to some other providers.
> a delay of at least 6-12 hours before the first byte of retrieval occurs
Huh. I wonder how it actually works behind the curtains. Do hey actually use HDDs to store all of that, or maybe is there some physical work involved in these 12 hours to retrieve a tape-drive from the archive…
Pro tip is to create a Google Workspace org and pay the $18/mo per 5TB of Drive storage to also get no egress fees. With rclone, the process is fairly simple especially when using a service account to authenticate (so no oauth revocations or refreshing).
The two work together well too - I prefer to let Restic back up to a separate local (magnetic) HDD for my repo, and then use rclone to sync the local repo to Backblaze B2. Compared to a direct cloud repo set up in Restic, it gives me much better restore times and lower costs for the common case where your local site is not completely destroyed (e.g. data corruption etc...), while still giving you the safety of a slower and more expensive off-site backup for when you really need it.
[1] https://github.com/geerlingguy/my-backup-plan