Here’s a typical sort of challenge: you’ve got a workload running in public cloud, but there’s a strong desire to get a copy of the workload on-premises. Maybe it’s something like an Oracle or SQL RDS export, or maybe it’s some other database you’d like to copy out.
Particularly for RDS protection, from a day to day basis you’re going to use cloud native protection — so something like Cloud Snapshot Manager, for instance. But what about that copy that’s going to land back on-premises, or maybe even go to another cloud? Your compliance copy matters, but so too do your recurring costs.
Well here’s where architecture matters when it comes to deduplication efficiency, yet again.
Let’s say you’re comparing Product X, who estimate their database backups will achieve 7:1 deduplication, and a Data Domain derived solution (e.g., DDVE in public Cloud — or for that matter an IaaS system in public cloud running a backup agent) that might instead give you 20:1 deduplication for database backups.
Now in either case, there’s going to be an export of some fashion, but then we need to either backup and replicate the data, or backup the data, and at some point that data is going to exit your cloud provider (in this case, looking at AWS) and come into your datacenter (or go to another cloud).
So let’s assume you’ve got 100TB of databases you need to get out each month. We talk about deduplication efficiency in terms of storage size, but what about data egress amounts?
Well, for X, at 7:1 deduplication, you’re going to need to pull 14.29 TB out each month. Using the AWS calculator, egressing that data via EC2 (i.e., a virtual machine holding the appropriate agents and tools) out of the Asia Pacific (Sydney) region is going to be a monthly fee of $2,262.94 (USD).
Via a Data Domain deduplication engine, at 20:1 deduplication efficiency? That’d just be 5TB per month egress, which for Sydney again would give you a monthly fee of $1,147.33 (USD).
Over the course of a year, that’s $27,155 USD vs $13,767 USD: a pricing increase of almost double due to inefficient deduplication approaches.
Architecture matters.