随笔分类 -  Parallel Computing

摘要：The thundering herd problem occurs when a large number of processes waiting for an event are woken whenthat event occurs, but only one process can pro... 阅读全文

摘要：One final hardware primitive is the fetch-and-add instruction, which atomically increments a value while returningthe old value at a partucular addres... 阅读全文

摘要：Some platforms provide a pair of instructions that work in concert to help build critical sections.On the MIPS architecture, for example, the load-lin... 阅读全文

摘要：Given our basic spin lock, we can now evaluate how effective it is along our previously describedaxes. The most important aspect of a lock is correctn... 阅读全文

摘要：Because disabling interrupts does not work on multiple processors, system designers started toinvent hardware support for locking. The earliest multip... 阅读全文

摘要：The name that the POSIX library uses for a lock is mutex, as it is used to provide mutual exclusionbetween threads, i.e., if one thread is in the crit... 阅读全文

摘要：From the introduction to concurrency, we saw one of the fundamental problems in concurrentprogramming: we could like to execute a series of instructio... 阅读全文

摘要：The other major component of any threads library, and certainly the case with POSIX threads, isthe presence of a condition variable. Condition variabl... 阅读全文

摘要：Beyond thread creation and join, probably the next most useful set of functions provided by thePOSIX threads library are those for providing mutual ex... 阅读全文

摘要：This chapter briefly covers the main properties of the thread API. Each part will be explainedfurther in the subsequent chapters, as we know how to us... 阅读全文

摘要：This chapter has set up the problem of concurrency as if only one type of interaction occursbetween threads, that of accessing shared variables and th... 阅读全文

摘要：One way to solve this problem would be to have more powerful instructions that, in a singlestep, did exactly whatever we needed done and thus removed ... 阅读全文

摘要：The simple thread example we showed above was useful in showing how threads arecreated and how they can run in different orders depending on how the s... 阅读全文

摘要：Let's say we wanted to run a program that created two threads, each of which was doingsome independently work, in this case printing "A" or "B". The c... 阅读全文

摘要：学习并行系统的性能可以用来挑选好的算法、评估硬件平台以及使用parallelism的收益。5.2.1 Execution Time5.2.3 Speedup当我们评估一个并行系统的时候，常常会考虑，如果让一个application并行运行的话，那么相对于sequential inplementation性能收益如何？这时会用到加速比(Speedup)这个概念，定义如下：Speedup is a measure that captures the relative benefit of solving a problem in parallel. It isdefined as the rati 阅读全文

摘要：Parallel Computing并行计算是同时利用多种计算资源来解决可计算问题：1.利用多个处理器。2.一个问题可以被分解为多个离散的、能并发解决的部分。3.每个离散部分可以被进一步分解为一系列指令。4.各个离散部分之间的指令可以同时在不同的处理器上执行。5.使用控制/协同机制。Concepts and Terminology1.Supercomputing and HPC(High Performance Computing)2.Node: comprised of multi-processors/CPUs/cores, memory, network interface etc.3. 阅读全文