[Yuval Dimnik]

Disclaimer: I do NOT work for AWS or any other cloud provider.
In addition, there is no argument about the fact that just storage – GB for GB, the cloud is more expensive.
That said, off the top of my head, here are some reasons S3 is great:

Variable OpEx vs. CapEx.
==============
See this for easier than any text: https://www.cloudtp.com/wp-content/uploads/2016/04/traditional-view.svg

Data Redundancy
==============
If you use QNAP with large drives and raid you risk raid rebuild failures and data loss. On the cloud – not your concern
https://www.enterprisestorageguide.com/raid-disk-rebuild-times

Logistics and logic
=============
Buy a QNAP, bring it on site, migrate, take it to your home and connect it to power/network while it takes some space. Need reasonable network at home to support data replication assuming anything is done at the office during the day.

vs.

Write your data to S3.
Done.

…Get
DR
A repository where you can share from to other users.
Work on your data using cloud services (e.g. https://aws.amazon.com/digital-media/aws-elemental/)
If cloud storage prices drop you enjoy that drop.
If you delete data, you don’t pay for that data anymore (unlike storage).
If a drive fails… Somebody else’s problem.

Flexibility, agility, the ability to adapt and change, adopt new technologies offered by the cloud etc are some additional benefits.

Yuval
yuval.dimnik@noobaa.com
https://www.noobaa.com

[Yuval Dimnik]

The cloud is not usable as an online storage, especially if your workload is not hosted on the cloud.
That said, it is a great solution for DR, archives and even nearline.
In these cases you won’t be exposed to the high read out of the cloud costs (egress costs) or metadata costs, and it’s an opex model on pay per use on data you actually store.

You can push 1TB of data per day with one dedicated line (you don’t want it to kill your email….) of 100Mb/s.
In the states, you can get 500Mb/s if you add $100/M (Not local so this is what I found https://www.verizon.com/home/fios-fastest-internet/) and then one can push 5TB per day.

The cloud providers provide a tool to import and export large amounts of data that can help with initial load and DR if the site was burned.
https://docs.aws.amazon.com/snowball/latest/ug/whatissnowball.html
https://cloud.google.com/data-transfer/docs/introduction

yuval.dimnik@noobaa.com
https://www.noobaa.com

[Yuval Dimnik]

“Once you take about 10 TB into consideration and, say, 2 years of storage then restore, the cost balloons into tens of thousands of dollars.”

Here are my calculations for the described scenario. Would love feedback.

Standard S3:
Storage costs: 10TB * $0.023 per GB (Standard Storage) = ~$230/M = $5,520/2Y
Egress costs: 10TB * $0.090 per GB (Data Transfer OUT From Amazon S3 To Internet) = $900
Total: $6,420/2Y

Glacier (cold archiving)
Storage costs: 10TB * $0.004 per GB (Glacier Storage) = ~$40/M = $96/2Y
Egress costs: 10TB * $0.090 per GB (Data Transfer OUT From Amazon S3 To Internet) = $900
Glacier Data Retrievals: 10TB* $0.0025 per GB (Bulk – 5 – 12 hours acceess time) = $25
Total: $996/2Y

https://aws.amazon.com/s3/pricing/

yuval.dimnik@noobaa.com
https://www.noobaa.com

[Yuval Dimnik]

When designing storage redundancy keep this question on your list:
“What is the process to get back to desired redundant state after failure?”

Unless you’re using SSD drives (which I would assume you do not) you should not use Raid-5/6.
You’ll waste capacity as a tradeoff to redundancy, but the chances to recover the Raid in failure scenarios is a bit low:
https://www.enterprisestorageguide.com/raid-disk-rebuild-times

I agree with Greg Janza’s benefits of Raid 10, but what will happen once a failure occurs?

An alternative is to look for redundancy through multiple copies or erasure coding for larger capacities.

Yuval
https://www.noobaa.com
yuval.dimnik@noobaa.com