The free lunch is over... but dinner has been ordered.

Multicore processors sound like the hit of the moment. However, these technologies are quite old. The first multicore processors had been designed on 2004 approximately. Over 10 years! That’s significant for a “modern” technology. By this, most people believe that computers can run much more programs in lower time. And their intuition could be right unless the fact that they don’t. Why this happens? The answer is simple: programs are not designed to be run on more than 1 processor at once.

This problem has created users’ disappointment because we, as users, want speed, effectiveness, reliability, and so many other characteristics from a computer. And this expectations increase on quantity and quality every year, due to marketing, and some other techniques sellers use to convince us to buy a brand-new computer every year (at least). 

As we can read on Suttler’s article, multicore processors have substituted the “Moore’s law”. Hardware engineers are not spending more time on researching how to increase clock speed because, no matter how simple or complex the processor architecture is, the processor dissipates more energy on heat proportionally to the clock speed. In other words, the faster the clock is, the more heat the processor dissipates. And, of course, heat can damage performance, even destroy the processor. This is one of the main reasons that clock speed is not a task to improve.

Nowadays, we use computers that have multicore processors. And this fact works hand-to-hand with hyperthreading and cache-size-increasing trends. As Suttler says: if we increase the size of cache blocks, we can avoid more “miss” events and change them for “hit” events. Then, the second technique in order to use all functionality given by multicore processors is: parallelism.

However, parallelism is not a common way to program, and this is due to its complexity. But, this is not an excuse to keep programing on single-core based way, because, when it comes to performance, the only way to speed up programs nowadays is by learning to parallelize our programs, and allow the multicore processors to use they physical advantages, where and when possible.

Programmers can keep programming under single-core layouts. But, if they want to have a better performance of their own codes, they must learn how to split processes, and exploit most of the resources of the new systems.

If we can manage 2 cores, there is a change that we can handle more cores, and give our users the possibility to do much more things at a time.

Notes:

A “hit” event occurs when the piece of data that the processor is waiting for is already loaded on cache.

A “miss” event occurs when the piece of data that the processor is waiting for is not located on cache, and the system must search for it on the principal memory or the hard drive, and then fetch it from that location.

Suttler, H (Aug, 2009). The Free Lunch is Over. Retrieved on August 23, 2015, from http://www.gotw.ca/publications/concurrency-ddj.htm