Could you test with mydumper ( https://github.com/maxbube/mydumper ), in order to differentiate between scaling via multiple connections dumping vs multiple servers?

Also, in my experience, parallelism in the dump part is less than half the problem, as restore from mysqldump typically scales much worse than creating the mysqldump.

Finally, the other problem that can occur is when, even though you have 100 tables, inevitably there will be one (or a few) table(s) that make up large percentage of the entire database size (e.g. an audit trail table or similar). mydumper seems to have features to help in this case, e.g. by chunking the tables into parts.

My point-of-view:
interesting
– when you don’t have advance storage capabilities, want to limit operational task impact on your production
– by using standard tool, architecture

But:
– doesn’t it make more complex ?
– implementation ?
– if you need to restore our master (user issue –> delete data –> replicated to other slave), you will need to restore your master and rebuild your replicated (DR more complex, RTO impacted)
– what is the cost on running multiple ‘host” and additional storage for (in this scenario) 3 replicas ?
– as you mention there is disk/Storage solution, certainly on modern one, they have advance feature can provide a other magnitude of service and in this context:
– snapshot : for TB, it takes seconds (only delta)
– clone : for TB, it takes seconds to have a clone of the running (real-time) master MySQL instance
– time efficiency: very fast
– space efficiency : for the read –> access the original block of the Master databases (shared resource). With your database of e.g. 1 TB you could have a clone with some MB/GB. You consume the same storage resource (cpu, disk, …) and to limit this you could do the clone from a older snapshot/backup.

Just to add few things as a suggestion ..

events and routines
to dump events , procedures and functions

also it would be great to have a hostname as part of each backup.

and last it would be great if there is a check for replication filters before execution of backup.