DATA CONSISTENCY Data processing must comply with the canons of consistency and consistency. When we process frequent information on a monoprocessor we load data into the cache that passes it to the CPU. It may happen that a valid transaction at the beginning can no longer be completed at the end, because while the usual steps were being performed, another transaction was terminated and the last unit was removed from the stock, for example, because it was a pipeline architecture process. where several sets of instructions are executed simultaneously but each goes in a different stage of advancement or a process type A consistency problem is generated if the stock cannot have negative quantities, and a roll back system must be initiated to maintain consistency. In multiprocessing this situation is even more complex and sometimes leads to drops or freezing of the operation, even if the software has contemplated the modus operandi. For example the sale of 120 . 000 tickets for daddy Yankee's recital fell apart in the first instance. A SIMD architecture team was enough, (Single Instructions, Multiple Data') because the operation is always the same. When the sales operation was repeated, the computer was changed, which was able to process the 120. 000 tickets on sale in about 60 . 000 purchases in the span of one hour, i.e. a thousand average tickets per minute. When the last entries of each category are processed, it happens that the purchase is accepted, but when it is processed, someone has already closed the purchase of the last seat, by using a transactional architecture type pipeline or a multiprocessor. If when reading the balance there were 60 left but another transaction left it at 56, the last transaction must be lowered from 56 and not from the last data read. If it was 2 and I'm selling 4 What do I do? Do I sell two? I reverse the operation? . These issues make coherence critical.