The next big leap in advanced computing technology gets a name

0

Small preview image

Back in June we reported about the research conducted by professor Uzi Vishkin at the University of Maryland’s A. James Clark School of Engineering, and what they claim to be the next generation of desktop computers. The emphasis is on parallel power and the fact that the new architecture will not make any significant changes to how software is written. Basically, the hardware handles the threading. At the same time they also announced a competition for finding a name for the new computer, and a few days ago they announced the winner and the name of the multi-processor solution; ParaLeap.



The person who came up with winning name was Jaryd Malbin, a 21-year-old student from Westport, Conn., attending Franklin and Marshall College in Lancaster, Pa.


Quoting the previous report from June, 30;


“They’ve namely developed a new organization that allows all of the processors to work together and at the same time allow the programmers to use the same kind of programming as with serial designs. One of the major reasons desktops have not become more parallel, but instead refined the serial design, is because of the software programming complexity a parallel, or even massively parallel, design involves, which is now a thing of the past if the prototype turn out to be just as good as they claim it to be.”


Professor Vishkin himself explains the approach like this:


“The following ABC story will help clarify the problem. Abraham (A) employs Benjamin (B) as a cleaning person for his house. B comes in alone once per week, works for several hours, and is paid by the hour by A. One day, A gets the following alternative offer from a cleaning company (C). C proposes to work at the same hourly rate as B, but is willing to provide A with 1000 cleaners for his house.


A is asked to give C a plan for how to clean the house (what to do first, second, and so on) using the 1000 cleaners. That is, each step of A’s plan can use up to 1000 cleaners simultaneously.  The only limitation is that A must give one plan to be used every week. The single plan should work efficiently, regardless of whether, in the previous week, A’s teenaged child had 100 people party in the basement, or A hosted a formal dinner, or had an open house for his friends, or spilled coffee on the bedroom carpet, etc.


The benefit of C’s offer for A could be pretty drastic as A will pay C the same hourly rate he has been paying to B, only C will use a workforce 1000 times larger than B’s. Note that theoretically this could save $999 out of every $1000 that A currently pays B. So A said yes to the C’s marketing person and signed an annual contract with them.


After signing the contract A was told that none of the 1000 cleaners speak his language, and that he must come up with a cleaning plan that will always work well without any further input or intervention from A. C will translate A’s plan to the language the cleaners understand once and they will execute it every week.”


In theory, ParaLeap could use as many as 1000 processor on a single PCB, but working design models use a more modest amount of 64 processor, which is still impressive though. More information about ParaLeap and Vishkin’s research can be found at www.ece.umd.edu.

Subscribe
Notifiera vid
0 Comments
äldsta
senaste flest röster
Inline Feedbacks
View all comments