Assignment 2: Multiprocessing

MSMP and SMP Master-slave multiprocessing, or MSMP, is when one CPU (the master) controls the system, all resources and scheduling. Only this master can execute the operating system. The other CPU’s are slaves that perform all of the work assigned by the master (Englander, 2009). This configuration is not widely used anymore. Conversely, Symmetrical multiprocessing, or SMP, is where each CPU has the same access to the operating system, all resources, and memory. In this set-up, each CPU is identical (Englander, 2009). In a master-slave configuration, the master holds the memory and is the one who allocates the needed memory to the other CPU’s. The problem with this is that the master CPU becomes bottlenecked. A solution of this could be to enable direct access memory (DMA). DMA is an operation in which data is copied from one resource to another resource in a computer system without the involvement of the CPU (TechTerms, 2013). An example would be data being sent directly from an attached device, like a flash drive, directly to the memory. The microprocessor does not have to do anything, and therefore speeds up the overall computer operation.
In SMP configuration, memory access could present a problem. All the memory access is posted to the same memory bus. If there are only a handful of CPU’s, this is fine. However, when there are larger numbers of CPU’s competing for access (anything over about 12), this could bottleneck and present a real problem. The solution to this is NUMA, which stands for Non-Uniform Memory Access. NUMA alleviates the bottlenecks by limiting the number of CPU’s on one memory bus. The CPU’s are set up into smaller systems called nodes. Each node has its own processors and memory, and is connected to the larger system through a high speed interconnector (Microsoft, 2013).
Regarding inputs and outputs (I/O), CPU’s in a symmetrical multiprocessing configuration have full access to all I/O devices. In the master-slave configuration, the master CPU will allow which CPU’s can have access to I/O devices, and when they can perform I/O operations.
Regarding instruction execution, the master-slave configuration has a master cache, two smaller direct mapped slave caches, and a slave instruction cache for supplying and executing instructions. In SMP, instructions sets are included on additional registers.
Hyperthreading
When work is assigned from among the available programs to be executed, the independent segments of those programs are called threads. Hyperthreading is when a CPU performs a parallel execution of two or more threads simultaneously, also referred to as simultaneous thread multiprocessing (STM) (Englander, 2009). Hyperthreading would not be able to be used in a master-slave configuration. This is because each slave has to wait until the master is free to get a new work assignment, and cannot be assigned more than one assignment. Conversely, hyperthreading is managed in a way that it very similar to symmetrical multiprocessing. Since there is not a dedicated processor for master tasks, and each CPU is independent, hyperthreading (STM) and SMP can be used together. The operating system in a SMP can schedule two processes at the same time, each on a different CPU. Therefore, less time is spent waiting as in the MSMP and more time working efficiently and executing the processes (Lui & Schmider, 2011). The best benefit of hyperthreading is that it allows the scheduling maximum flexibility to fill execution slots, thereby making the most efficient use of available resources. The system that uses hyperthreading uses a fraction of the resources and therefore limits the amount of resource waste (Stokes, 2002).
Applications
Even though the master-slave configuration is not the most widely used anymore, there are still some application for which MSMP is still best suited for. Some of these include: game controllers, finance applications, biology and physics applications. This is because reactions and responses are made to every event that occurs. Any application that is based on simulation and modeling would be best suited for the master-slave configuration. Symmetric multiprocessing configurations are used in a wide variety, and is the most commonly used. The most widely used is mobile technology. The Android devices are built around SMP technology, with multiple CPU’s that do different things. For example, the newest smartphones are Quad Core. One CPU may be used to run applications, while others are used to run various hardware. The programs running cannot use the main memory to communicate with each other (Android, 2013). Some of the benefits that users partake of with SMD on the mobile devices include: * Faster web page load times; * Lower power consumption and higher performance per watt; * Higher quality game play experience; * Highly responsive; * Faster multitasking.

References
Android (2013). SMP primer for Android. Retrieved from http://developer.android.com/training/articles/smp.html.
Englander, I. (2009). The Architecture of computer hardware, systems software & networking. John Wiley & Sons: Hoboken, NJ.
Lui, G. & Schmider, H. (2011). The double-layer master-slave model: A hybrid approach to parallel programming for multicore clusters. High Performance Computing Virtual Laboratory, Queen’s University. Retrieved from http://www.hpcvl.org/sites/default/files/dmsm/DMSMmanual.pdf.
Microsoft (2013). NUMA support. Retrieved from http://msdn.microsoft.com/en-us/library/windows/desktop/aa363804%28v=vs.85%29.aspx.
Stokes, J. (2002). Introduction to multithreading, superthreading, and hyperthreading. ArsTechnica. Retrieved from http://arstechnica.com/features/2002/10/hyperthreading/1/.
TechTerms (2013). DMA. Retrieved from http://www.techterms.com/definition/dma.