MULTIPROCESSING the ability of a computer to execute several programs simultaneously, such as a billing system program and a redirection program of a missile that carries satellites to their geostationary orbits. When the term was created in the 60s the input speed (reading telex tape and then 80-column punched cards) was much lower compared to the CPU speed. The temporary storage memory was then segmented in order to process several programs in parallel. In addition, a cache was used where the most repetitive and frequent lines of code were kept "at hand" to speed up the process, instead of loading the executable program from the hard disk whenever they are required, very common with routines or calculation subroutines. This memory is faster and is often today hosted on the same chip as the CPU. The architecture of computers has evolved to make them faster and faster. Initially it was improved with 4 methods. The first, to incorporate into the hardware a being of frequently used instructions (80/20, 20% of the instructions are used 80% of the time), which was defined as microprogramming and CISC architecture (complex instructions ser computer) typical of the IBM /360 series and later. With the popularization of high-level languages such as Cobol, C, 4GL this set becomes less necessary and computers with RISC (reduced instructions set computer) architecture are built. Now, regardless of the architectures, there are 3 ways to make a computer faster: 1 . Increase the speed of the clock, that is, accelerate the cycle of electrical pulses. This brings complexities of saturation, overlap, noises, signal echoes. This is solved by making the chip smaller and smaller, leading to manufacturing problems. 2 . Pipelining execution in parallel of the different tasks that an instruction requires (retrieve it from memory, decode it, retrieve the operands, execute it and store the results). These tasks can be done as in pipes in parallel, so that when task 3 of an instruction is executed, task 2 of the next instruction is executed simultaneously, and task 1 of the instruction following the next. This technique is highly developed and all modern chips use it intensively and exhaustively. 3 . The third way of optimization is that of superscalar, which is the execution of several instructions at once in a single CPU cycle. This idea allows to multiply by three or by four the effective speed of the CPU. It brings the complication that when there are forks in the code, series of instructions are no longer required and must be rolled back. This makes executing 6 simultaneous instructions uneconomical. When it has come to squeeze in extreme the lemon has been derived in a fourth technology, the multiprocessing. Over the years, MIMD technology computers have been developed that have 2 and up to 16 CPUs that work simultaneously, parallel and independently, with the same addressable main memory (physical memory vs . virtual memory) for all . This creates the need for complex consistency (return of last recorded value) and consistency (when to deliver the last value) protocols. There are today multiprocessors with distributed or shared memory, which would already constitute several independent processors, but integrated.