Often I hear people tailing of the “network” as the “computer” and the Cloud as its implementation.
Now I happened to read a few data on supercomputers that I’d like to share with you and that convinced me of the enormous distance between what can be done by a Cloud and what can be done by a supercomputer.
In the Cloud you can have an enormous processing power by adding as many computers as you please. However this processing power cannot be equated to speed of processing. It is like saying that the Amazon river has a tremendous water transport capacity, although the “speed” of the water it transport is a few meter per second. If your goal is to get a water stream at 100 meter per second the Amazon is not a solution.
Now, going back to the Cloud and the Supercomputer.
The IBM Roadrunner (2008) has a 1 PFlop capacity, that is it can perform 1 million billion floating point instructions per second. It is not a single processing unit but a cluster of units (as it is the case for any supercomputers today), so you cannot equate capacity and speed. However it ha 50 thousands channels, each working at 5GBps, connecting the various processors. In a Cloud you have channels with a connection speed that is at least 5,000 times slower.
If you take Blue Waters, the most recent IBM supercomputer delivered this year, it has a ten times higher capacity (10 PFlops) and 5 million optical cables each running at 10 GBps. By 2014 supercomputers will expect to reach 1 EFlops (1 billion billions floating point instructions per second) and optical connections will total 1 billion at a channel data rate approaching 50 GBps. Interesting to note that the cost of the optical interconnection in today Blue Water is 10% of the total supercomputer cost, whilst in the EFlops supercomputer it is estimated to be 40% of the total.
This kind of speed, infrastructure and even more important pace of evolution is a world apart from the Cloud. The two will live parallel life with no one superseding the other.