مقاله انگلیسی رایگان در مورد طراحی جلسات آزمایشگاهی با تمرکز بر پردازنده های واقعی – الزویر ۲۰۱۸

Most electrical and engineering schools around the world offer two or three courses on computer organization and computer architecture topics. These courses usually comprise both conventional lectures at classroom and practical sessions at laboratories. Computer architecture courses are typically considered as difficult courses by students mainly due to the wide range of topics that are covered as well as the intrinsic difficulty of some of them. In addition, topics are usually studied from a theoretical perspective, which discourages many students from continuing their education in computer architecture. Lab sessions are an excellent way to reinforce the theoretical concepts taught at conventional lectures. They provide a clear understanding about how the computer mechanisms studied at lectures work, which helps correcting any possible misunderstanding and motivates these students that might feel discouraged at classrooms. To this end, labs use computer simulation frameworks like Multi2sim [21] or Snipper [6], which model complex processors and their structures in detail. Working on small fragments of the simulator source code helps students to appreciate the details about how specific processor mechanisms work, allowing them to acquire a sound knowledge about internal architectural mechanisms from a practical perspective. Therefore, simulators play an important role in post-graduate courses, especially when a major goal of the course is to provide research skills to students. While simulators are valuable tools to study the details of hardware, they fail to provide an overview of how the distinct components of the machine interact to each other. For example, how last level cache (LLC) misses impact processor performance. One reason that explains this drawback is that simulating a current multicore with complex cores and a huge cache hierarchy is time consuming. In fact, running just a millisecond of execution on real hardware can take several hours with the simulator. To deal with such problems, we propose lab sessions where students work on real hardware. The proposed labs make use of the hardware performance counters implemented in most recent processors from the major manufacturers Intel [14], AMD [9], IBM [18] or ARM [2]. Performance counters consist of a set of special purpose registers that allow tracking advanced processor events such as committed instructions, run cycles, memory accesses, or branch misses, among many others. In summary, this paper presents a new approach to study computer architecture topics at lab sessions focusing on real hardware and makes two main contributions.