On 9/14/05, Fawad Lateef <fawadlateef@xxxxxxxxx> wrote: > On 9/14/05, Raghu Vadapalli <iprsvp@xxxxxxxxx> wrote: > > On 9/14/05, Andre Luiz da Silva <andrack@xxxxxxxxx> wrote: > > > Hello Friends! > > > > > > I'm studying the kernel memory management in the Linux Kernel, with the > > > book: > > > "Understanding the Linux Virtual Memory Manager", but I don't understading > > > the > > > "fundamental" direferences between NUMA and UMA memory architeture. > > > > > > I perform a serach at the Google and the Kernel Documentation > > > (Documentation/vm/numa), > > > but this make me very confused... > > > > > > Do Somebody has any explanation about the concepts of NUMA and UMA memory > > > architeture? > > > > > > > > > Thanks!!! > > > > > JUST FIXING MY TOP POSTING...BEFORE SOMEONE POINTS IT OUT. > > AFAIU, > > > > UMA --- CPU to Memory bank access times are same. > > > > NUMA -- CPU to memory bank access times are different for different cpus. > > > > So you always want to allocate from the memory bank which is near to you. > > > > correct me if I am wrong. > > Raghu > > > > UMA has Uniform Memeory Access time, means memory is shared between > several processors and NUMA is Non-Uniform Memory Access in which > processors are assigned their own memory banks which are closer to it > and have less access time !!!! Here I m attaching a gif to make it > clear ........... > > Definations from google search for NUMA: > > NUMA (non-uniform memory access) is a method of configuring a cluster > of microprocessor in a multiprocessing system so that they can share > memory locally, improving performance and the ability of the system to > be expanded. NUMA is used in a symmetric multiprocessing (SMP) system. > An SMP system is a "tightly-coupled," "share everything" system in > which multiple processors working under a single operating system > access each other's memory over a common bus or "interconnect" path. > Ordinarily, a limitation of SMP is that as microprocessors are added, > the shared bus or data path get overloaded and becomes a performance > bottleneck > > Non-Uniform Memory Architecture. Usually on an SMP system, all memory > beyond the caches costs an equal amount to reach for each CPU. In NUMA > systems, some memory can be accessed more quickly than other parts. > Later kernels have some NUMA support for preferring to use memory > "nearer" to the CPU, rather than higher-latency distant memory > > Sorry forgot to attach the gif ..... now attaching .... -- Fawad Lateef
Attachment:
NUMA-UMA.gif.gz
Description: GNU Zip compressed data