Parallel processing!

March 8, 2020

Why Parallel Processing?.

Gearman Job Server is the first parallel/distributed processing framework that I’ve learned/worked on which is indeed a very old framework out there. One of the things I like about Gearman is that it’s very generic and you can play around a lot. In the last post I wrote how to setup the gearman job server and in this post I’m going to brief you why we need parallel processing with some examples:

Example 1: Divide and conquer.

Let’s say we have 10000 colored balls having red, green and blue colors and 3 buckets of each color. The task is to segregate the balls to the respective colored buckets.

1 Approach:

You pick a ball, look at the color, put it inside the same colored bucket and repeat. Time Taken: Assuming T to be the time taken to segregate one ball, total time would be (10000*T).

2 Approach:

Call K number of friends if possibles, Get K number of beers, Divide the balls among y’all so that each one of you will process 10000/K number of balls only, Everyone follow the I Approach. Time Taken: Assuming T to be the time taken to segregate one ball, total time would be (10000*T/K).

Example 2: Resource Optimizion.

Again, Let’s say we have 10000 balls and 3 buckets A, B, C with A having clean water, B having soap water, C having clean water. The task is to clean all the balls pun intended :D.

1 Approach:

You pick a ball, put it inside A, then B and then C and repeat for all the balls.

NOTE: when you are using one bucket, others are not in use or idle.

2 Approach:

Call your two best friends Alex and Bob, Assign each task of using bucket A, bucket B, bucket C to y’all. Now what you will do is just pick a ball, dip it in bucket A and give it to Alex and repeat. Alex will dip it in bucket B and will pass the ball to the Bob and repeat. Bob will dip it in bucket C and Voilla, One ball is cleaned. Time Taken: Assuming T/3 to be the time taken to process a ball through one bucket, but now there’s a catch

NOTE: when you are using one bucket, others are being used as well. After everty T/3 time, one ball is getting cleaned which makes the effective time taken to clean a ball is T/3 +- some error;

total time would be ~ (10000*T/3).

So, I think I’ve convinced you that parallel processing is gooood.

NOTE: Parallel processing can only be done with some specific types of tasks. Some tasks are inherently sequential. A fine example would be making a cup of tea, you have to follow the steps. One way to identify if a task can be parallelized or not is that if there is a some sort of repetition or lack of system resource utilization (like in example 2) then you can think of making the task into sub tasks and processing them parallelly.

How Gearman helps you achieve that?.

In the above examples we saw how we can do tasks in parallel ways. Let’s say we have a task which can be divided into K subtasks. To complete the task what we can do in gearman is create a worker which will process one subtask at a time. Now just instantiate P number of workers and submit all the K tasks to the gearman job server which will keep them in a queue. What you will see is that K number of subtasks are being processed parallelly i.e. every worker is processing one subtask. Hurrray!! We just processed our task parallelly!!