The time to run P is represented by Tseq(P). The times to run P1 and P2 are represented by Tseq(P1) and Tseq(P2). It is obvious that, when running on a single node, we can observe two things:
- P2 cannot start running before P1 is complete. This is represented by P1 -- > P2
- Tseq(P) = Tseq(P1) + Tseq(P2)
Let's assume that P overall takes 10 minutes to run on a single node. Out of these 10 minutes, P1 takes 2 minutes to run and P2 takes 8 minutes to run on a single node. This is shown in the following diagram:
Now the important thing to note is that P1 is sequential in nature. We cannot make it faster by making it parallel. On the other hand, P2 can easily be split into parallel subtasks that can run in parallel. So, we can make it run faster by running it in parallel.
The major benefit of using cloud computing is the availability of a large pool of resources and many of them are used in parallel. The plan to use these resources for a given problem is called an execution plan. Amdahl's law is used comprehensively to identify the bottlenecks for a given problem and a pool of resources.