Mon, Apr 11, 2005 at 12:06 AM
Haven’t been writing for a while, discovering something to
revolutionizing the world isn’t easy.
So my concept of today is about computer cluster. I had this idea
since I just learn about the concept of collective computing. There is
something truly backward about that: Why we must be using computer of
identical specification to do cluster? Why couldn’t we do it like the
way of distributed computing which utilize many different computers of
different specification and different hardware?
One reason maybe because it is easy to do cluster with all the
identical computers. It wouldn’t be a technological challenge to
distribute a task to identical node. But doing so require customer to
buy many of identical hardware, which is not always available for
every type of customer.
What is the solution? Euler’s concept of asymmetry cluster. In such
an architecture, the master program decide what is the best level of
organization of the specific hardware/software component of each
computer. It could either be complete P2P, or a P2P with supernodes,
or single dictatorship, or group-equal dictatorship, or group-unequal
dictatorship. Of course, before that there is a process of checking
out the parameter of each nodes, and test-run.(They are designed by
programmers beforehand) And a calculation would take place.
After it has decide which aspect of performance of the cluster would
be optimize relative to current task according to human input. It then
implement the structure by sending out agents. Agents then temporarily
take over each computer as the given rank. It is ready to receive the
order from various level(s) of master.
How to deal with the problem of asymmetry in computing power? Use
the Mathematic of L.C.M and H.C.F. For instance, if it learn that
machine A take 10 cycle to complete a given task (From general
knowledge input by human), and machine B take 5 cycle to complete a
task. Then it will arrange the task in such a way that machine A would
not need the result of machine B every 5 cycle. And of course, this
program also need to sort out the logical relatedness of individual
task. It is as simple as when step k require a variable to pass from
step k-1, then either step k must be carried via one machine. Or step
k-1 would be finished when another machine is ready to take step k.
The Mathematic of this wouldn’t be more difficult than the task
I named as Euler’s adaptive cluster. It adapt to situation rather
than vice versa. With the computing power we have now, it isn’t
different to do the calculation and implement a scheme like this.