dual processors and linux 1766
wizzywiz Sounds like an open-ended homework buttignment to me. Some of the issues I would be concerned about are (I do...
Actually, the original loop was mostly CPU bound, because essentially all the disk access were very quickly cached. The gzip command above is running while I am typing this response...
wizzywiz
"... Runqueues and load balancing
Linux 2.6 is able to efficiently manage more processors than Linux 2.4, the previous stable version. The ability to deal with different amounts of computational resources, from small embedded CPUs to mbuttive 64-processor supercomputers, is called "scaling;" it is one of the new kernel's many strengths. One might think that a scheduler designed to manage one CPU could never be adapted to managing 64 CPUs; however, the Linux 2.6 scheduler can manage large multiprocessor systems through the use of special lists called "runqueues." A runqueue stores information about the processes running on a single CPU; there is one runqueue for each CPU in the system. The information contained in the runqueues allows the processor to seamlessly transfer control from one CPU to another, using a method called "load balancing."
That is almost certainly the case. (However, my test case used Mandriva Linux 2006, not a Mac.) I recall being informed in either around 1987 and maybe around 1980 that Unix adjusted a process's...
Load balancing is a way to ensure that no CPU's resources are going to waste while another CPU is overstressed. If the scheduler finds that one runqueue has many more processes in it than another, one or more processes may be moved from the larger runqueue to the smaller one. The load balancer is invoked whenever a runqueue is empty; if no runqueues are empty it is invoked periodically. The periodic timer allows the system to maintain a reasonable load balance across many CPUs without devoting too much time to moving processes from one CPU to another. Balancing the load whenever a runqueue is empty allows the scheduler to ensure that precious CPU power never goes to waste. There is one exception to this load balancing rule - some special processes may be fixed to a certain runqueue. This attribute is called "thread affinity;" a thread is simply another name for a process.
It is important to note that if SMP (symmetric multiprocessing) support is not enabled when one compiles the kernel, none of the load-balancing code will be enabled and no time is wasted trying to balance the load on a single CPU..."