elapsed time from submission to the completion of execution. (In other words, the time that we must wait before execution starts becomes significant.)
Via considering 300 seconds startup-time and 100 seconds execution time of a benchmark on EC2 (32 VMs) vs. a propability of about only 1/3 that the super computer finishes the task within 400 s after job submission, Ian comes to the result that for such things EC2 is more convenient, which is — of course — true, but only regarding these very special conditions.
We have to consider that EC2 is billed per hour, which means that it’s uneconomical to run EC2 instances for less than an hour. Hence, the comparison made has no particular meaning for real use cases. Instead of running 32 nodes for 100 seconds each, one should have started one VM which would have finished within the first hour for 1/32 of the total price. EC2 is simply not adequate for tasks that need less time than an hour. Considering at-least-one-hour-lasting-tasks changes the whole examination dramatically, since then “the time that we must wait before execution starts becomes” less significant compared to the execution time: the super computer will take lead again, because of its computing power.
Finally, while answering the question What’s faster — a supercomputer or EC2?, one has to say that most of the poeple are not that lucky that they’ve access to a super computer. But they have access to EC2, instantly! I think that is the real advantage of EC2 :-).