Is this possible. (Distributive computing type question)

[ICRS]

In simplest terms the alogithm is this:

Select an initial Q1, R1, A1, B1, C1

For n = 1 to 1 Million ( Some very large number)
{
Yn = fy(Qn)
Zn = fz(Qn, Rn)
Xn = fx(Yn, Zn, An,Bn, Cn)
An+1 = fa(Xn)
Bn+1 = fb(Xn)
Cn+1 = fc(Xn)
Qn+1 = fq(Yn,Zn)
Rn+1 = fr(Zn)
}


For complex equations, or when the number of variables increases it takes a very long time to run on 1 computer, some times even over 24 hours. So I thought if I split it up into 2 or 3 computers that run through it together simultaneously then I can make it quicker. This will be similar to the idea of distributive computing.

 

[degibson]

What you've specified here is a long serial chain of computation -- aside from individual functions that happen to have no explicit data dependence (e.g. fy and fz), there is very little parallelism. The problem is that there is a dataflow order constraint on, say, Rn+1, such that Rn must be computed before Rn+1 for all n.

In order to distribute this application to multiple threads/machines/computers/whatever, you will probably have to find a way to express [ABCQR]n+1 as functions that don't have a loop carry dependence on [ABCQRXYZ]n. That may not be an easy thing to do, but seeing as I don't know what the functions do, thats the best help I can offer.