Second AWS SimuLearn course, Cloud First Steps. The scenario is an island's stabilization system, currently running on a single point of failure, that needs to move to something that survives a hardware or facility-level outage without going down.
The obvious fix sounds simple: run it twice. Two EC2 instances instead of one. But "run it twice" only buys you anything if the two copies can't both go down for the same reason at the same time, and that condition is doing more work than it sounds like.
What actually makes two things independent
An AWS Availability Zone is a physically separate facility: its own power, its own cooling, its own networking, isolated from the other AZs in that region. Two EC2 instances in the same AZ are still, underneath the redundancy, sharing a single physical failure domain. A power event or a facility-level hardware failure takes both of them down together, and "redundant" turns out to have meant nothing.
Two instances in two different AZs don't share that failure domain. Whatever takes one down, a rack failure, a power issue, anything scoped to that facility, has no path to reach the other one. That's the actual mechanism behind "high availability" in this scenario, not a bigger number of servers, but servers that fail independently of each other.
The lab makes this concrete rather than abstract: launch one EC2 instance, configure it, confirm it's actually running by having it display its own instance details in a browser. Then do it again, deliberately in a different Availability Zone within the same region. Two instances that look identical from the outside, but only one of the two configurations is actually redundant.
The storage side needed the same discipline, in a different shape
The computational modules attached to each instance needed storage that behaved like a local disk, continuous block-level reads and writes, not files retrieved as complete objects over HTTP. That ruled out S3 for this specific role even though S3 is the right answer elsewhere in AWS's own SimuLearn catalog (the beach-conditions static site from the earlier course is exactly S3's use case). Amazon EBS attaches block storage directly to an instance, which is what "the modules read and write to it as part of normal operation" actually requires.
Two different problems in this one scenario, and two different failures to avoid: assuming proximity counts as separation on the compute side, and assuming any "reliable storage" service is interchangeable on the data side. Neither mistake shows up until you ask what specifically the requirement depends on, not just what the requirement is generally about.
Loading article