Since using auto-scaling, it seemed to be no different from clustering except that the scale is automatically changed.
I think what they have in common is that both increase availability by distributing traffic across multiple instances.
The difference is that auto-scaling can increase and decrease instances.
I want to know if my concept is correct or not.
I would also appreciate it if you could attach a link for reference.
No. Auto scaling by itself, the compute instances are not required to be any type of cluster. Rather, it is a capacity management feature. An optional component of an infrastructure.
Cluster is a multiple node system that closely coordinates access to resources they share. Distributed storage systems. Multiple node hypervisors managing VMs. Failover clusters. Stacked network switches. Shared memory supercomputers. Each of these has nodes that talk to the others about who has what.
Contrast to a counter example of an auto scaling group where each node is independent. Consider a test configuration, where instances start CPU burning tasks. The stress program, or computing a trillion digits of pi. After exceeding a CPU use threshold, more instances would be started. But they didn't talk to each other. Perhaps the system collected metrics like CPU and memory utilization to assess capacity, but I do not consider that in the same category as a failover cluster.
Add a load balancer in front of this group of compute instances, and sure the combined system can scale out automatically and be highly available.
