They didn't. They just tried to store data on EBS volumes, you know, like they are supposed to. But EBS performance is very bad, and incredibly variable. So they would end up getting timeouts trying to write to the volumes, which broke their DB replication, which took the site offline for hours to re-replicate.