4 ms·

At any time an instance may become un-responsive and so will local storage, no? I don't see how this helps?

• joevandyk 15y ago
  postgresql replication to other instances? i believe the point is that instance storage is more reliable than ebs.
  • jjm 15y ago
    Why would you want to take a chance on that? Every instance should be thought of as expendable. They can go down at anytime. EBS is really just like any other NAS/iScsi vol and it wouldn't be such a problem if they did what they're supposed to. That is, be consistent in read, write, and durability.
    • joevandyk 15y ago
      What do you mean by "take a chance on that"? If the instance goes down, you can failover quickly to another one that you've been replicating to.
      • jjm 15y ago
        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. Check http://joyeur.com/2011/04/22/on-cascading-failures-and-amazons-elastic-block-store/ http://joyeur.com/2011/04/22/on-cascading-failures-and-amazo...