|
The material in this section is for background only, and is NOT strictly relevant to specific topic of distributed memory programming. You may safely skim or skip this section without cause for worry that important material has been missed.
Tremendously exciting things are being explored into being in labs around the world. Some are already in the development stage, while others have barely made it out of left-field and into serious investigation. The following are just a few of the directions that are being explored, any of which could have astounding implications for computational performance.
- Optical Computing
Relays originally where electro-mechanical: a binary operation was performed when an electrical signal caused a mechanical switch to flip from one state to another. This was tremendously overshadowed by the current state-of-the-art, solid-state electronics, where the entire operation is electro-magnetic in nature. However, this is not the ultimate in switching speed: electrical devices retain their own built-in limitations, for example resistance, and the associated waste-heat product that naturally goes along with their operation, all of which works to limit the speed with which state-changes can be effected.
Optical computing refers to switching technology based completely on characteristics of light, most commonly its polarity. The light, of course, is coherent, or lasers, and the speed with which a change in the polarity of laser-light can be detected is considerably faster than that of the change-in-state of an electrical device, and doesn't carry with it the unwanted baggage of resistance and waste-heat.
- Atomic- and Molecular-Computing
Current computational devices (transistors and their cousins) are much, much smaller today than even a few years ago, but efforts are underway to develop effective devices at both the molecular and even atomic levels. Certain kinds of molecules have multiple conformations, or shapes, that they can be coerced into assuming, and the switch between the two shapes can be accomplished very rapidly. The different shapes can be discriminated very easily and also very quickly, and the decrease in functional-unit size is several orders of magnitude, meaning that many, many times the current number of transistors could be attained were molecular-level components able to be efficiently utilized.
Atomic-scale computing takes this down another few orders of magnitude, and is accomplished, in one approach, by being able to move individual atoms between a pair of locations, one signifying the "0" binary state, the other the "1" state. Individual atoms have been able to be deliberately placed for some years now; finding a way to do so quickly, and over many atoms at once, will enable a scale of computational construction that simply boggles the mind: supercomputing capabilities on postage-stamp sized computers, for example.
- Holographic Memory
Holography, the construction of 3-dimensional images by constructive interference of lasers, holds tremendous promise for being able to serve as a data storage medium of awesome potential, both in terms of capacity and in terms of retrieval. Multiple data objects can effectively be stored using the same area of storage, simply by using different incidence angles for the data beams, and each can be individually recovered by using its own unique beam angle for the read laser, and at optical speeds rather than electrical.
- Biological Computing
Very recently, a researcher used the base-pairs of the DNA strands as simple binary building blocks, out of which he constructed the initial conditions relevant to the Traveling Salesman Problem, a well-known problem in optimization. Using well-established replication techniques, he created a huge number of "initial conditions", and then added the necessary ingredients allowing the DNA strands to find their complements ... in this case, a true complement constitutes a solution to the problem. There are understandably a huge number of solutions, but the real problem is to find the shortest solution, which translates in this context to one of weight: the most optimal DNA solution will be that solution which weighs the least.
Assuming that your particular problem could be coded into this form, the entire operation going from presentation of initial conditions to acquisition of a range of solutions could take anywhere from hours to days to weeks, and possibly even longer -- so why get excited? Because the number of solutions is truly astounding, and is so much larger than the time it took to get them, that there is a thousand-fold speedup over the number of solutions found by normal computer-oriented methods. For example, assuming that the DNA operations themselves take between 10 minutes and 4 hours to complete, that means that they are anywhere from 10^11 to 10^12 times slower than the fastest supercomputers...but there are 10^20 DNA strands (each one comparable to a compute-node in a massively parallel computer), as compared to the 10^1 to 10^4 nodes in MPP systems, all of which yields a 10^3 to 10^4 scale edge of DNA computing over standard electronic computing.
This style of computation is so new it can't even be said to be in its infancy, and the opportunities for further acceleration and sophistication are still unexplored, although potentially staggering. For example, a recent article ("Will DNA Compute?", High Performance Computing and Communications Week, May 4, 1995) claims:
Molecular computers should be able to operate 1,000 times faster than the current 100-billion-operations-per-second speed of electronic supercomputers, with an energy efficiency that is one billions times greater, and an ability to store data in about one trillionth the space required by existing storage media such as video tape.
- Multi-Frequency Communications
Communications links have characteristically utilized a single frequency for data transfer; multiple signals were multiplexed, either in terms of time ("during this period, this signal is being served, then this one, then ...")
or by phase ("run this signal with 0 phase change, this one at an offset of 20 degrees, ...").
New technology is now capable of putting multiple independent frequencies into the same media (fiber), and boosts data-capacity tremendously.
|
