If you accept that any of your instances can go down at any time, you accept the fact all of them can be down at one (the same) time.
You might think your safe with reserved instances too, thinking you've reserved the dedicated time with EC2. Well what happens when the entire network stack goes down or block storage the reserved instance rack depends on goes down? So does all 10 or 20 reserved instances you had up, at once.
Also, it was this very replication/snapshot mirroring feature of EBS that cascaded into (network,etc) congestion.
You might think your safe with reserved instances too, thinking you've reserved the dedicated time with EC2. Well what happens when the entire network stack goes down or block storage the reserved instance rack depends on goes down? So does all 10 or 20 reserved instances you had up, at once.
Also, it was this very replication/snapshot mirroring feature of EBS that cascaded into (network,etc) congestion.
Check http://joyeur.com/2011/04/22/on-cascading-failures-and-amazo...