DEV Community

Discussion on: Cloud: IO limits gone full circle

Collapse
 
databasesponge profile image
MetaDave 🇪🇺

Yes, we recently hit a limit on RDS IO burst with AWS RDS PostgreSQL, and then another limit on EBS IO burst.

It was very confusing, but unlike in the bad old days when every single database was weirdly IO bound and always would be, it only took a 15 minute outage and a bit of money to scale to a new instance size and type.

Reminiscing about the old magnetic storage RAID days, you could change the IO escalator on a RAID array and achieve close to the theoretical maximum of throughput for a database with large IOs (e.g. data warehouses). Good times.

Collapse
 
fritshooglandyugabyte profile image
Frits Hoogland • Edited

Thank you for your reply Dave.

I can't tell if it was an IO burst on 'your' side, or a problem with burst limits. I think bursting is the 21st century snake oil, but that is another discussion.

I agree that cloud means improvements have been made in many area's, which were impossible in the old days. I love 'infrastructure as code', and vividly remember the days where adding infrastructural components meant something had to be ordered and physically arrived at the data centre, and then had to be assembled, installed and configured.

The goal of the article is to warn that despite the many advantages that cloud gave us, there are things that have come back and moved in the opposite direction.

I have first hand seen a cloud vendor proclaiming they provided unlimited capacity. That doesn't exist.