خلاصة:
The last few years have seen an explosion in the deployment and use of the
Internet, networking and telecommunication technologies. This was
followed by significant increases in the speed and capacity of computing, for
example Petaflop supercomputers are becoming common. We will examine
some of the developments; explain their importance and potential impact.
Many forecasts and predictions have been made about the impact of the
increases of computing capacity and the growth of the Internet and the world
wide web. In this talk we will introduce some of the favorite predictions and
will analyze the possibilities for their realization in the long run. The
analysis shows that there exist hard limits on the growth of the Internet and
the increase in computing capacity. They prove that it is unlikely that some
of the predictions will hold in the long run. The restrictions are based on
basic physical and economic limitations, which generate tight bounds on the
realization of such predictions. The bounds will occur much faster than
expected by the simple forecasters.
ملخص الجهاز:
Table 1 displays the maximal problem size (number of cities) solved by a complete enumeration for different computer capacities (instructions per second) and different elapsed times devoted to the solution process.
This is a pessimistic limit considering that at present, top super computers can execute instructions at a rate of over a Petaflop (10'") Bailey (Bailey, 1997) instructions per second, putting the effective size of data independent TSP problems that can be solved today to optimality in the range of 18 to 21 cities (depending on the amount of elapsed time allocated to the solution procedure).
Table 2 shows the lower bound on the amount of energy required when we assume that Planck constant is used to probe/flip a bit and the computer performs / instructions per second.
Today's top of the line supercomputers (the Roadrunner) (Shannon , Murphy , Niemier , Izaguirre & Kogge ; Grice,2009; Haviv,2006) perform a Petaflop of operations per second and consume 2 to 4 Megawatts to achieve that level of computing capacity (Grice,2009).
Table 2: A lower bound on the amount of energy consumed by computers of different capacities using Planck's constant Number of operations per .
Table 4 lists the problem sizes that can be solved using the dynamic programming based algorithm, for different computing capacities and elapsed times, the table assumes that each function evaluation during the solution process requires the execution of 10 instructions and that 40 bit level operations have to be executed for each instruction.