|
Re: parallel processing
One somewhat cool tool that never seemed to make it out of Microsoft in house development was DMAKE, a distributed make utility for large multiple source projects. The network contained a computer with the source files, another computer with the DMAKE app and associated tools, and multiple computers that were compiler servers. DMAKE would send off messages to each of the compiler servers to start compiling individual source modules. When a compile server signaled completion, DMAKE would then assign yet another module to be compliled, until all modules were compiled. Once the compile steps for a library module were completed, DMAKE would then start a link server to link all the just produced objects into a library or directly to an executable if it was a single library project. I saw it demoed while taking some device driver class for Windows NT at Microsoft University (back in the 1990's). The compiler and linker servers were kept effeciently busy until all the libraries were made and then linked to create a completed project. I don't recall if DMAKE kept a history of compile times from previous builds in order to optimize the scheduling. Computers are so fast these days that it would take a huge project for something like DMAKE to make sense now, but it was impressive at the time.
Another example in the parallel algorithm thread was distributing the analysis of satellite photos by breaking them up into multiple parts and sending each part to a different computer to operate on.
|
