Thursday 9 January 2014

SUMMARY (First and Second Lecture)

LETS BEGIN J

Hello fellas!!

We will be posting the summary of each lecture and consequently keep you updated about everything deliberated upon in the discussion. Kindly go through the posts and comment in case of any amendments or additional points that we might have overlooked.

Also, we might have messed up with the names or missed out on a few people but that is purely due to human error.

Thank you.

BASIC IDEA


-> INTRO

  Parallel processing is division of a task into subtasks which are handled by multiple processors simultaneously and later their outputs are combined to give the final result efficiently as mentioned by Rishul and Sweety.

-> MULTITASKING

  Various tasks are performed simultaneously in the multi core processors.

-> BETTER EFFICIENCY

  It is mainly for faster execution i.e higher throughput.

-> ANALOGY

  It is analogous to a Project team in a company wherein there are departments like Finance, Marketing, Designing etc which perform distinct tasks independently but towards a common goal as mentioned by Sunchit.

-> HIGHER THROUGHPUT

  As the number of processors increase, work load on the system increases but work done per unit time also increases as mentioned by Vikas Mulodhia.

-> PERFORMANCE STANDARDS

  In parallel processing, certain performance standards should be maintained in terms of execution time and data handling as mentioned by Varun and Sweety.

-> INTERRUPT HANDLING

  Interrupts coming from I/O devices may be divided among the processors based on their priorities as mentioned by Vikas Mulodhia and Vishal.

-> HARDWARE-SOFTWARE TRADE OFF

  Hardware has high speed and low cost and software has low speed and high cost so there's a trade off so H/W or S/W implementation depends on the requirements as mentioned by Vishal Kumar.

-> SPACE TIME BARGAIN

  As the number of processors increase, execution time reduces but the space(memory) requirements increase. Greater the speed achieved, more will be the memory requirement for handling the system as mentioned by Tusshar.

-> MANAGING THE MULTIPLE PROCESSORS (AN OVERHEAD)

  In parallel processing, there needs to be a system which allocates tasks to the processors and manages the resources. Though parallel processing speeds up the process but there is an overhead of coordinating and managing the tasks handled by these processors as mentioned by Uzma.

-> MULTI CORE PROCESSING

  Multicore processing is a subset of Parallel processing as mentioned by Rishabh Jain.

-> USER BASED SYSTEM MANAGEMENT

  The user may decide when to run which application and how much memory to allocate to each of them as mentioned by Suraj.

-> DATA PARTITIONING

  Data might be partitioned and processed by multiple processors for faster computations as mentioned by Saloni.

-> INTER PROCESSOR COMMUNICATION

  Processors need a proper means of Communication and Data Sharing so as to ensure proper processing of partitioned data as mentioned by Riva.

-> SCHEDULING

  There needs to be a scheduler which decides in what sequence jobs are allocated to the processors.

-> SYNCHRONISATION

  All the jobs of the parallel processors need to be synchronised so that no processor remains idle and data is processed reliably as mentioned.

-> CONCEPT OF HETEROGENEOUS PROCESSORS

  All the processors may handle different types of jobs and may have different set of registers and other resources.

-> I/O HANDLING

  I/O devices might be handled together by multiple processors as mentioned by Sunil.   

-> VLSI Technology:

  The hardware requirement for parallel processing is provided by the VLSI technology which uses the various logic families. VLSI Technology refers to the Silicon technology being used extensively in the ICs.


OVERHEADS

 

-> CONFLICT RESOLUTION

  Two or more processors might try to access the same portions of memory at a given time. Such conflicts need to be resolved.

-> INCONSISTENCY

  Multiple processors might edit or manipulate the same data leading to inconsistency. Such a situation also needs proper handling.


POINTS TO PONDER UPON

-> MEMORY CONCERN

  In parallel processing, multiple processors may use parts of a common memory or individual memories as designed.

-> ENVIRONMENTAL CONCERN

  Parallel processing requires more hardware so its power requirements would be more and heat loss might be high too. So we need to think about its effect on the environment too.

-> APPLICATIONS

  Various field in which parallel processing is a necessity.



IMPORTANT POINTS

 

-> Every processor is a Von Neumann processor which has its own memory.

-> Interrupts are program associated i.e a program decides which devices will be used and which interrupts are to be generated.

EXECUTION TIME OF PROGRAM IN A UNIPROCESSOR SYSTEM

FACTORS AFFECTING EXECUTION TIME:

1. No of Instructions (N)

2. Cycles per instruction (CPI)

  CLOCK CYCLES PER INSTRUCTION = CYCLES FOR DECODE N EXECUTE + CYCLES FOR FETCH

  Fetch time = Time for decode and execute + K * Memory Access time

  where K = (Speed of memory / Speed of processor)

3. Time period of one clock cycle (T)

  This depends on the VLSI technology involved in making the circuits.


So, EXECUTION TIME OF A PROGRAM = N * CPI * T

 

No comments:

Post a Comment