Sie sind auf Seite 1von 2

Muhammad Arslan Munir

12-NTU-4007

In Multithreading or parallel, multiple operations perform in single step. Many form of parallel computing such as instruction level, data, bit level, and task parallelism. The basic idea of parallel algorithm is, its divide the main problem to sub-problem thats may be solved parallel by multi processor. In parallel computing model used are Shared-Address-Space and Massage-Passing. In Shared-Address-Space global memory accessible to all computers like work station and In MassagePassing, each process give it own block memory like Intel paragon. The model of parallel Algorithm is Parallel Random Access Machine (PRAM). It is theoretical model for parallel machine. Its use as the same way as RAM use in sequential algorithms for time complexity. PRAM is uniform memory access means it can access by any processor at any address in same time. There are many common models for PRAM. Exclusive-read Exclusive-write (EREW) o In which every memory cell can be read and written only one processor at a time. Exclusive-read Concurrent-write (ERCW)

Concurrent-read Exclusive write (CREW) o In which multiple processor can reads memory at a time but only one can read. Concurrent-read Concurrent-write (CRCW) o In which multiple processor can read as well as write the memory same time. There are some common terms in parallel algorithms. Concurrent Process o In which program divided into multiple process which are run on single processor Distributed Processing o A program is divided on multiple processes which run on multiple machines. These machines usually connected by Local Area Network. Parallel Processing o In which programs divide in multiple processes which run on multiple processor. These processors may be in one machine, execute one program at one time and had high communication. There are some issues in parallel programming. Breaking down the problems in smaller problem than its run in parallel by allocating different processors and also define sequence to run this task. Some processes must wait at determined points for results from other processes. Some performance Issues in parallel Algorithms are. Load should be balance in parallel Algorithms means all processors have same amount of work and avoiding the idle processor. Parallel Algorithms should be scalable means use more process to run a task fast and do same thing in large task. Some global control in parallel Algorithms are. Single instruction Single data (SISD) o In which involves sequential computing

Muhammad Arslan Munir

12-NTU-4007

Multiple instruction Single data (MISD) o On same input every processor can do different things like shape detecting in picture Single instruction Multiples data (SIMD) o Every processor do same thing on different data input. Multiple instruction Multiple data (MIMD) o Different inputs different data.

Das könnte Ihnen auch gefallen